Crack anomaly detection system and method based on image recognition

Abstract

The invention relates to the technical field of image recognition, in particular to a crack anomaly detection system and method based on image recognition. The method comprises the following steps: acquiring a hydropower station dam monitoring image; performing surface crack analysis based on the hydropower station dam monitoring image to obtain surface crack data; performing crack type division according to the surface crack data to obtain mesh crack data and transverse crack data; identifying the crack width according to the transverse crack data to obtain crack width data; performing width abrupt change anomaly detection based on the crack width data to obtain width abrupt change anomaly data; and performing communication link interruption analysis based on the width abrupt change abnormal data to obtain communication link interruption data. According to the method, the accuracy and real-time response capability of crack anomaly detection are improved based on the image recognition technology, and the scientificity and reliability of crack dynamic evolution and risk early warning are enhanced.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a crack anomaly detection system and method based on image recognition. Background Technology

[0002] Traditional crack anomaly detection methods have weak crack type differentiation capabilities, relying heavily on basic image processing algorithms such as edge detection or grayscale thresholding. These methods cannot effectively identify and differentiate between different types of cracks, such as mesh cracks, transverse cracks, and longitudinal cracks, leading to unclear data for subsequent crack development trend analysis and maintenance decisions. Furthermore, there is a lack of effective fusion between image recognition data and crack monitoring sensor data, failing to establish a dynamic correspondence between high-frequency visual information and low-frequency physical quantity data. Dynamic tracking of crack anomaly trends is also impossible; most methods can only identify static crack morphology at a specific moment, lacking the ability to monitor time-series changes such as crack propagation rate and width abrupt changes, making it difficult to provide timely warnings of potential structural risks. Finally, image recognition data cannot be effectively integrated with sensor data... The precise spatial mapping of sensor placement locations leads to a disconnect between visually abnormal areas and monitoring point data. Furthermore, when sensor data anomalies occur (such as data drift, communication interruption, or failure), the lack of a high-frequency image data verification mechanism makes it difficult to determine whether the anomaly originates from the actual evolution of the crack itself. Typically independent of the structural response analysis system, image recognition results are not combined with physical mechanisms such as material fatigue and stress concentration, making it impossible to determine whether the crack has posed a substantial threat to the dam structure. When monitoring data is abnormal or lost, the system lacks a mechanism to trace the cause of communication link failures, making it impossible to determine whether the data interruption is caused by equipment overheating, image transmission delay, or sensor malfunction, thus affecting the system's robustness and operational efficiency. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide an image recognition-based crack anomaly detection method to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, an image recognition-based crack anomaly detection method includes the following steps: Step S1: Acquire monitoring images of the hydropower station dam; perform surface crack analysis based on the monitoring images of the hydropower station dam to obtain surface crack data; classify crack types according to the surface crack data to obtain network crack data and transverse crack data; Step S2: Identify the crack width based on the transverse crack data to obtain crack width data; perform width abrupt change anomaly detection based on the crack width data to obtain width abrupt change anomaly data; perform communication link interruption analysis based on the width abrupt change anomaly data to obtain communication link interruption data; Step S3: Perform crack anomalous growth analysis based on the network crack data to obtain crack anomalous growth data; perform material fatigue analysis based on the crack anomalous growth data to obtain material fatigue data; perform dam component fracture detection based on the material fatigue data to obtain dam component fracture data. Step S4: Calculate the communication link interruption time based on the communication link interruption data; collect crack morphology data based on the communication link interruption time; predict the crack propagation direction based on the dam component fracture data; transmit the crack morphology data and crack propagation direction to the crack display system to perform the abnormal crack display task.

[0005] This invention achieves precise identification and classification of surface cracks by acquiring and analyzing monitoring images of hydropower station dams. This helps refine crack characteristic data, improves the accuracy of crack type classification, and provides a reliable data foundation for subsequent crack width identification and anomaly detection. Dynamic detection of crack width and analysis of abrupt width changes enable timely detection of abnormal crack propagation trends and sudden changes, enhancing the sensitivity and timeliness of anomaly identification. Combined with communication link interruption analysis, the communication status of the monitoring system is effectively monitored, ensuring the continuity and integrity of data acquisition and reducing monitoring blind spots caused by communication failures. The combination of abnormal crack growth and material fatigue analysis deepens the understanding of crack development mechanisms and the degradation process of dam material performance, helping to identify potential structural risks early and improving the accuracy and early warning capabilities of fracture detection. Simultaneously, based on communication link interruption time statistics and crack morphology acquisition, the dynamic evolution process of cracks can be accurately reconstructed, assisting in the prediction of crack propagation direction and providing a scientific basis for risk assessment and maintenance decisions. Finally, the crack morphology data and propagation direction are integrated and transmitted to the crack display system, realizing the visualization and multi-level management of crack anomaly information, enhancing users' intuitive understanding of crack anomalies and response efficiency, and significantly improving the intelligence level and application value of the overall monitoring system.

[0006] Preferably, step S1 specifically includes: Step S11: Acquire monitoring images of the hydropower station dam; Step S12: Detect crack edges based on monitoring images of the hydropower station dam to obtain crack edge data; Step S13: Perform contour repair based on crack edge data to obtain surface crack data; Step S14: Identify network cracks based on surface crack data to obtain network crack data; Step S15: Identify transverse cracks based on surface crack data to obtain transverse crack data.

[0007] This invention acquires dam monitoring images and performs crack edge detection, accurately extracting crack edge features and improving the precision and reliability of crack detection. Combined with contour restoration technology, it effectively fills in missing parts of crack images, ensuring the integrity and continuity of crack data, thereby improving the accuracy of subsequent crack morphology analysis. By identifying network cracks and transverse cracks separately from surface crack data, the classification of crack types is refined, helping to more comprehensively reflect the morphological characteristics and development status of cracks, providing rich and high-quality basic data support for multimodal data fusion and deep anomaly detection. This method breaks through the limitations of traditional systems with a single data source, achieving multi-dimensional and precise identification of crack morphology, significantly improving the accuracy and robustness of crack anomaly identification, effectively reducing the risk of false alarms caused by environmental interference and equipment failure, and laying a solid data foundation for accurate prediction of crack propagation trends and structural safety risk assessment, meeting the needs of modern intelligent monitoring systems for comprehensive, accurate, and intelligent data processing.

[0008] Preferably, step S14 specifically includes: Step S141: Extract crack centerline information based on surface crack data; Step S142: Identify crack intersection relationships based on crack centerline information; Step S143: Calculate the crack connectivity based on the identified crack intersection relationships; Step S144: Predict the crack network trend based on crack connectivity; Step S145: Mark the network crack region based on the crack networking trend; Step S146: Calculate the crack density based on the network crack region; perform high-density crack region clustering based on the crack density to obtain network crack data.

[0009] This invention achieves precise capture of the main structure of cracks by extracting crack centerline information, which helps to deeply understand the geometric morphology and evolution of cracks. Identifying crack intersections based on crack centerlines reveals the interconnections and influences between cracks, providing crucial data support for crack network structure analysis. Calculating crack connectivity quantifies the degree of crack interconnection, providing a scientific basis for assessing the complexity and potential expansion risks of crack networks. Predicting crack network trends helps to identify the trend of cracks evolving from isolated to networked structures in advance, enabling timely preventative and maintenance measures. Marking network crack regions and calculating crack density allows for precise location of high-risk crack areas. Combined with cluster analysis of high-density crack regions, it effectively identifies key areas where cracks are concentrated, providing targeted guidance for subsequent structural safety assessments and crack remediation. These techniques overcome the limitations of traditional single-data analysis, achieving multi-dimensional and multi-level dynamic analysis of crack morphology and risk identification, significantly improving the accuracy and intelligence level of crack anomaly detection, and laying a solid foundation for intelligent and precise monitoring and maintenance of cracks in hydropower dams.

[0010] Preferably, step S15 specifically includes: Step S151: Calculate the crack orientation angle based on the surface crack data to obtain the crack orientation angle data; Step S152: Determine suspected horizontally distributed cracks based on crack orientation angle data, and obtain suspected horizontally distributed crack data; Step S153: Detect stress concentration areas in the dam body based on suspected horizontally distributed crack data; Step S154: Conduct structural expansion detection based on the stress concentration area of ​​the dam body to obtain structural expansion data; Step S155: Determine the failure of steel reinforcement anchorage based on structural expansion data, and obtain steel reinforcement anchorage failure data; Step S156: Identify transverse cracks based on the rebar anchorage failure data to obtain transverse crack data.

[0011] This invention achieves accurate identification of crack spatial distribution direction by calculating crack orientation angles, enhancing the analytical capability of crack geometry. Based on crack orientation angle data, it filters out suspected horizontally distributed cracks, facilitating the focused monitoring and analysis of crack types closely related to the dam structure's stress. By detecting stress concentration areas in the dam, it can promptly identify potential structural anomalies, providing crucial evidence for preventing structural damage. Based on stress concentration areas, it detects structural expansion, accurately capturing local deformation and expansion of the dam, reflecting changes in structural stability. By judging rebar anchorage failure, it reveals the internal rebar anchorage status, helping to identify structural reinforcement needs in advance and prevent further deterioration. Finally, it utilizes rebar anchorage failure data to assist in transverse crack identification, improving the accuracy and specificity of crack type classification. These collaborative techniques overcome the limitations of traditional single-data and static analysis, achieving multi-dimensional deep integration analysis of structural stress and crack behavior, significantly improving the accuracy and intelligence level of crack anomaly detection, and providing scientific and comprehensive technical support for the safety monitoring and maintenance of hydropower dams.

[0012] Preferably, step S2 specifically includes: Step S21: Identify the crack width based on the transverse crack data to obtain the crack width data; Step S22: Calculate the width gradient change value based on the crack width data; Step S23: Identify abrupt slope points based on the width gradient change value; Step S24: Extract width mutation anomaly data based on mutation slope points; Step S25: Calculate the width mutation timestamp based on the width mutation anomaly data; collect camera monitoring logs based on the width mutation timestamps; calculate the camera overheating status based on the camera monitoring logs to obtain camera overheating status data; Step S26: Detect continuous data packet loss based on camera overheating status data to obtain continuous data packet loss data; perform communication link interruption analysis based on continuous data packet loss data to obtain communication link interruption data.

[0013] This invention, by identifying crack width and calculating its gradient change value, can accurately capture subtle trends in crack width changes, effectively identifying abrupt slope points in crack width and accurately extracting and locating width abrupt anomalies. Based on the statistical timestamps of width abrupt anomaly data, combined with camera monitoring logs, it can promptly identify the operating status of monitoring equipment at critical moments of crack changes, especially camera overheating, providing important evidence for equipment failure early warning. Furthermore, by using camera overheating data for continuous packet loss detection, it enhances the ability to identify communication link interruptions, ensuring the continuity and integrity of monitoring data. Overall, it achieves multi-dimensional linkage analysis of crack width changes, equipment status, and communication link anomalies, effectively compensating for the monitoring blind spots caused by the limitations of single sensors and static thresholds in traditional systems. It significantly improves the accuracy and robustness of anomaly detection, enhances the comprehensive perception of crack development dynamics and the health status of monitoring equipment, and provides more comprehensive and reliable technical support for the safety monitoring and maintenance of hydropower dams.

[0014] Preferably, step S3 specifically includes: Step S31: Calculate the crack propagation rate based on the network crack data; perform crack anomalous growth analysis based on the crack propagation rate to obtain crack anomalous growth data; Step S32: Identify the stress concentration region of the crack based on the abnormal crack growth data; Step S33: Calculate the dynamic load based on the crack stress concentration region to obtain dynamic load data; calculate the cyclic stress amplitude based on the dynamic load data; perform crack propagation simulation based on the cyclic stress amplitude to obtain crack propagation data; Step S34: Predict the material fatigue life based on crack propagation data to obtain material fatigue data; Step S35: Conduct fracture detection of dam components based on material fatigue data to obtain fracture data of dam components.

[0015] This invention achieves dynamic analysis of abnormal crack growth by calculating crack propagation rate, enabling timely capture of rapid crack development trends and improving sensitivity and accuracy in identifying abnormal crack changes. Identifying stress concentration areas based on abnormal growth data helps accurately locate weak points in the structure, providing key areas for subsequent monitoring and reinforcement. Combining dynamic load data to calculate cyclic stress amplitude and conduct crack propagation simulations allows for in-depth analysis of crack evolution under complex loads, enhancing the scientific rigor and reliability of crack development trend prediction. Material fatigue life prediction enables timely assessment of the safety status of dam components, providing early warning of potential fracture risks and effectively ensuring the stability and safety of hydropower dam structures. The overall method integrates multi-source data and multi-dimensional physical parameters, overcoming the limitations of traditional systems relying on static thresholds and single data sources. This improves the accuracy and robustness of anomaly detection, mitigates the impact of environmental interference and equipment failure, supports more comprehensive dynamic crack monitoring and risk assessment, and significantly enhances the practicality and application value of intelligent monitoring systems.

[0016] Preferably, step S35 specifically includes: Step 351: Calculate the cumulative fatigue damage value based on material fatigue data; Step 352: Identify highly fatigued areas of dam components based on cumulative fatigue damage values; Step 353: Acquire acoustic emission signals based on the high fatigue region of dam components; perform signal preprocessing based on the acoustic emission signals to obtain the acoustic emission signals to be processed; calculate the peak amplitude based on the acoustic emission signals to be processed; calculate the signal energy based on the peak amplitude. Step 354: Determine the high-frequency acoustic emission signal energy region based on the signal energy; conduct strength degradation detection of dam component materials based on the high-frequency acoustic emission signal energy region to obtain dam component material degradation data; Step 355: Based on the material degradation data of dam components, conduct fracture detection of dam components to obtain fracture data of dam components.

[0017] This invention achieves a quantitative assessment of the fatigue state of dam components by calculating cumulative fatigue damage values, accurately identifying highly fatigued areas and providing a scientific basis for key monitoring and maintenance. Based on the acquisition and preprocessing of acoustic emission signals from highly fatigued areas, combined with the calculation of peak amplitude and signal energy, it enhances the sensitivity to micro-cracks and damage within the material, improving the accuracy and reliability of signal analysis. By calibrating high acoustic emission signal energy areas, it precisely locates the parts of material strength degradation, enabling early detection of material performance decline in dam components. Combining material degradation data with component fracture detection allows for a more comprehensive and timely understanding of the dam structure's safety status, supporting early warning and scientific maintenance decisions. The overall solution integrates multi-source monitoring data and multi-dimensional physical characteristics, overcoming the limitations of traditional reliance on single sensors and static thresholds. It significantly improves the accuracy and robustness of anomaly detection, reduces the impact of environmental interference and equipment failure on monitoring results, ensures the integrity and effectiveness of data acquisition, and enhances the system's ability to deeply mine and apply the dynamic evolution laws of cracks, meeting the high requirements of modern intelligent crack monitoring systems for real-time performance, accuracy, and comprehensiveness.

[0018] Preferably, step S355 specifically includes: Calculate tensile strength based on the material degradation data of dam components; The fatigue limit reduction factor is determined based on tensile strength. The fatigue zone of dam components is determined based on the fatigue limit reduction factor. A three-dimensional model of dam components was constructed based on the fatigue region of dam components. High-temperature simulation was performed on the three-dimensional model of the dam components to obtain high-temperature data of the dam components; Microscopic damage data was obtained by detecting microscopic damage based on high-temperature data of dam components; A thermally induced damage evolution map was drawn based on microscopic damage data; The evolution of thermally induced cracks was analyzed based on the thermally induced damage evolution map to obtain thermally induced crack data; Fracture prediction of dam components is performed based on thermally induced crack data, and fracture data of dam components are obtained.

[0019] This invention calculates tensile strength based on the material degradation data of dam components, achieving a precise quantitative assessment of the material's mechanical properties and laying the foundation for subsequent fatigue performance analysis. Using tensile strength to determine the fatigue limit reduction factor scientifically reflects the changing trend of material properties with fatigue damage, improving the accuracy of fatigue limit determination. Combining the fatigue limit reduction factor with the calibration of fatigue regions enables precise positioning of critical damaged areas of the dam, guiding key monitoring and maintenance. A three-mode model is constructed based on the fatigue region, comprehensively considering multimodal response characteristics to enhance the comprehensiveness and precision of structural behavior simulation. High-temperature simulation is conducted using the three-mode model. This system obtains temperature distribution and thermal response data of components under high temperatures, providing a basis for analyzing thermally induced damage. Based on high-temperature data, it detects microscopic damage, revealing the internal thermal damage mechanism of materials and improving the depth and accuracy of damage identification. By drawing thermally induced damage evolution maps, it visually displays the spatial distribution and development law of thermally induced damage, assisting in dynamic monitoring and trend prediction. Based on the thermally induced damage evolution, it performs thermal crack evolution analysis, accurately depicting the crack generation and propagation process, and improving the predictive ability of crack evolution. Combined with thermally induced crack data, it performs fracture prediction, effectively predicting the failure risk of components and achieving early warning and risk management. The overall solution achieves deep fusion and dynamic modeling of multi-source, multi-dimensional physical parameters, overcoming the limitations of traditional single-data and static analysis, significantly improving the accuracy, robustness, and practical value of crack monitoring, and meeting the high requirements of modern intelligent monitoring systems for comprehensiveness, accuracy, and real-time performance.

[0020] Preferably, step S4 specifically includes: Step S41: Calculate the communication link interruption time based on the communication link interruption data; Step S42: Acquire abnormal crack images based on the communication link interruption time; Step S43: Calculate the crack length based on the abnormal crack image; calculate the crack curvature based on the abnormal crack image; Step S44: Integrate crack length and crack curvature to obtain crack morphology data; Step S45: Predict the direction of crack propagation based on the fracture data of dam components; Step S46: Transmit the crack morphology data and crack propagation direction to the crack display system to perform the abnormal crack display task.

[0021] This invention achieves precise quantification of data transmission anomalies by statistically analyzing communication link interruption times, facilitating timely detection and location of communication faults and ensuring the data integrity and continuity of the monitoring system. It acquires abnormal crack images based on interruption times, ensuring timely capture of crack states at critical moments and avoiding monitoring blind spots and data loss. Calculating crack length and curvature using abnormal crack images comprehensively characterizes the geometric features of cracks, improving the accuracy and detail of crack morphology analysis. Integrating crack length and curvature information generates complete crack morphology data, providing rich foundational data support for subsequent analysis of crack dynamic evolution and propagation direction. Predicting crack propagation direction by combining dam component fracture data enhances the scientific judgment of future crack development trends, helping to formulate targeted maintenance and repair strategies in advance. Transmitting crack morphology data and propagation direction to the crack display system enables intuitive visualization of crack anomaly information, facilitating multi-level and multi-angle understanding of crack development and improving the interactivity and response speed of monitoring results. The overall solution effectively addresses the shortcomings of traditional systems that rely on a single data source and static threshold analysis, improves the ability to fuse multi-source data and identify dynamic anomalies, enhances the accuracy, robustness, and real-time response of monitoring, and meets the high standards of modern intelligent crack monitoring systems for data integrity, refined analysis, and visualization.

[0022] Preferably, this specification also provides an image recognition-based crack anomaly detection system for performing the image recognition-based crack anomaly detection method described above. The image recognition-based crack anomaly detection system includes: The crack type classification module is used to acquire monitoring images of hydropower station dams; perform surface crack analysis based on the monitoring images of hydropower station dams to obtain surface crack data; classify crack types based on surface crack data to obtain network crack data and transverse crack data; The communication link interruption analysis module is used to identify the crack width based on the transverse crack data to obtain crack width data; to perform width abrupt change anomaly detection based on the crack width data to obtain width abrupt change anomaly data; and to perform communication link interruption analysis based on the width abrupt change anomaly data to obtain communication link interruption data. The dam component fracture detection module is used to perform crack abnormal growth analysis based on network crack data to obtain crack abnormal growth data; perform material fatigue analysis based on crack abnormal growth data to obtain material fatigue data; and perform dam component fracture detection based on material fatigue data to obtain dam component fracture data. The crack display module is used to count the communication link interruption time based on the communication link interruption data; collect crack morphology based on the communication link interruption time to obtain crack morphology data; predict the crack propagation direction based on the fracture data of dam components; and transmit the crack morphology data and crack propagation direction to the crack display system to perform abnormal crack display tasks.

[0023] The present invention relates to an image recognition-based crack anomaly detection system. This system can implement any of the image recognition-based crack anomaly detection methods of the present invention. It serves as a medium for coordinating the operation and signal transmission between various modules to complete the image recognition-based crack anomaly detection method. The internal modules of the system cooperate with each other to improve the accuracy and real-time response capability of crack anomaly detection, and enhance the scientificity and reliability of crack dynamic evolution and risk warning. Attached Figure Description

[0024] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the steps of the crack anomaly detection method based on image recognition of the present invention; Figure 2 This is a detailed flowchart of step S1 in the present invention; Figure 3 This is a detailed flowchart of step S14 in the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] The technical method of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0026] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0027] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0028] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a crack anomaly detection method based on image recognition, the method comprising the following steps: Step S1: Acquire monitoring images of the hydropower station dam; perform surface crack analysis based on the monitoring images of the hydropower station dam to obtain surface crack data; classify crack types according to the surface crack data to obtain network crack data and transverse crack data; In this embodiment, high-definition industrial cameras deployed at key locations on the hydropower station dam are used to acquire real-time monitoring images of the dam surface. The cameras used have a resolution of at least 4000×3000 pixels, and the acquisition frequency is once every 10 minutes. The image format is lossless RAW compression to ensure complete image details. The acquired raw images undergo preprocessing, specifically including image denoising (using median filtering with a filter window size of 3×3 pixels) and grayscale normalization (linearly mapping pixel grayscale values ​​to the range of 0-255) to eliminate the influence of illumination variations. Subsequently, the edges of cracks on the dam surface are extracted using the Canny operator (high threshold 0.3, low threshold 0.1) based on an edge detection algorithm. The extracted crack contours are then segmented into fracture segments using connected component analysis, with the minimum connected region set to 50 pixels to filter out noise points. Crack type classification relies on crack orientation and morphological features. Network cracks are defined as crack groups with more than three branches and branch angles all less than 45 degrees. Transverse cracks are defined as cracks whose main orientation forms an angle greater than 75 degrees with the longitudinal direction perpendicular to the dam and whose length is greater than 500 mm. Type classification is achieved using geometric morphology methods based on thresholds, outputting corresponding network crack and transverse crack data. The data structure includes crack initiation and termination coordinates, length, width, and orientation angle.

[0029] Step S2: Identify the crack width based on the transverse crack data to obtain crack width data; perform width abrupt change anomaly detection based on the crack width data to obtain width abrupt change anomaly data; perform communication link interruption analysis based on the width abrupt change anomaly data to obtain communication link interruption data; In this embodiment, the crack width is calculated based on image grayscale profile analysis of the transverse crack data obtained in step S1. Specifically, the pixel grayscale variation curves on the crack outline are extracted. Combined with image resolution (each pixel corresponds to a physical length of 0.1 mm), grayscale values ​​are scanned in the vertical direction of the crack outline. A dual-threshold method is used to distinguish between crack and non-crack areas. The low threshold is set to 50 to characterize low-grayscale areas within the crack, and the high threshold is set to 150 to determine the crack boundary. The width is calculated as the physical length of the grayscale range below the 50 threshold. By comparing the crack width data collected continuously at multiple times, the crack width mutation is calculated using a difference method, and a width mutation anomaly threshold of 5 mm is set. Crack segments exceeding the threshold are identified as width mutation anomaly areas. Based on the width mutation anomaly data, combined with real-time monitoring logs of the communication link, the data transmission interruption time is statistically analyzed. The link interruption time is statistically analyzed by collecting packet loss rate and signal strength indicators in the communication system. A packet loss rate exceeding 10% and a signal strength below -85 dBm are defined as a communication link interruption. The duration is statistically analyzed with a precision of seconds, and the start and end times of the interruption are recorded.

[0030] Step S3: Perform crack anomalous growth analysis based on the network crack data to obtain crack anomalous growth data; perform material fatigue analysis based on the crack anomalous growth data to obtain material fatigue data; perform dam component fracture detection based on the material fatigue data to obtain dam component fracture data. In this embodiment, the network crack data obtained in step S1 is used to calculate abnormal crack growth data. The specific method includes time series analysis, calculating the crack growth rate by continuously collecting the rate of change of network crack length and crack branch number, and using linear regression to fit the crack growth curve. The abnormal growth threshold is set to a growth rate exceeding 0.2 mm / day. Based on the abnormal crack growth data, material fatigue analysis is performed. The material fatigue analysis uses a fatigue accumulation calculation method based on Miner's rule. Input parameters include crack growth rate, cyclic load amplitude (based on historical load sensor data, amplitude range 0~100 MPa), and fatigue life curve (determined by material testing, fatigue limit is 45 MPa, SN curve slope is -3). The cumulative fatigue damage value D is calculated, D=∑(n_i / N_i), where n_i is the actual number of cycles, and N_i is the allowable number of cycles under the corresponding stress amplitude. Based on the cumulative fatigue value of the material and combined with the stress intensity factor K in fracture mechanics, the fracture criterion K≥K_IC is used for fracture detection of dam components. K_IC is the material fracture toughness parameter, with a value of 85 MPa·√m. The fracture detection data includes the fracture location coordinates and the fracture risk level.

[0031] Step S4: Calculate the communication link interruption time based on the communication link interruption data; collect crack morphology data based on the communication link interruption time; predict the crack propagation direction based on the dam component fracture data; transmit the crack morphology data and crack propagation direction to the crack display system to perform the abnormal crack display task.

[0032] In this embodiment, based on the communication link interruption time counted in step S2, crack morphology images are re-acquired using a configured high-definition camera within the corresponding time period. The acquisition process requires image acquisition to be completed within 5 minutes of link restoration to ensure data timeliness. Crack morphology data acquisition includes the calculation of crack length and crack curvature. Length measurement is based on the length of the crack skeleton line in the image, converted to physical distance using pixel counting (scale 0.1 mm / pixel). Crack curvature is calculated using the three-point curvature method on the crack skeleton line. Specifically, three consecutive points P1, P2, and P3 on the crack skeleton are selected, and the corresponding curvature κ = 2S / (L1) is calculated. L2 L3), where S is the area of ​​the triangle, and L1, L2, and L3 are the lengths of the three sides. Statistical analysis is performed on all skeleton points to output crack curvature distribution data. Based on the dam component fracture data in step S3, the crack propagation direction is derived by combining fracture mechanics directions. Specifically, the maximum principal stress direction method is used, and the principal stress direction is calculated using the material stress tensor to determine the crack propagation trend along this direction. Finally, the crack morphology data and crack propagation direction are uploaded to the crack display system via a dedicated data transmission protocol. The display system is configured to support multi-dimensional data visualization, overlaying crack image with crack propagation direction arrows, supporting real-time refresh, and using the GeoJSON standard data format to ensure compatibility and efficiency of data interaction and display.

[0033] Preferably, step S1 specifically includes: Step S11: Acquire monitoring images of the hydropower station dam; In this embodiment, industrial-grade high-definition cameras installed at key locations on the dam are used to acquire monitoring images. The cameras employ CMOS sensors with a resolution of at least 4000×3000 pixels, and the image format is RAW uncompressed to avoid loss of image detail. The cameras are equipped with automatic exposure and white balance adjustment functions, with a fixed exposure time of 1 / 100 second and an ISO sensitivity set to 400 to balance image brightness and noise. The acquisition frequency is set to automatically trigger acquisition every 10 minutes for continuous monitoring. Acquired images are transmitted via fiber optic network to a local server for storage and subsequent processing, using TCP to ensure data integrity. During acquisition, the lens is ensured to be perpendicularly aligned with the dam surface, with an installation angle deviation not exceeding 5 degrees to avoid perspective distortion. Environmental condition monitoring equipment simultaneously acquires temperature and humidity data for subsequent image preprocessing and correction.

[0034] Step S12: Detect crack edges based on monitoring images of the hydropower station dam to obtain crack edge data; In this embodiment, the original image acquired in step S11 undergoes crack edge detection processing. First, the image is preprocessed by using a 3×3 median filter to remove salt-and-pepper noise, followed by histogram equalization to stretch the grayscale range to 0-255, enhancing the contrast between the crack and the background. Edge detection is performed based on the Canny operator, with specific parameters set to a low threshold of 0.1 and a high threshold of 0.3. The standard Sobel operator is used to calculate the image gradient direction. Edge connectivity is achieved through 8-connected region analysis, with a lower limit of 50 pixels for the connected region area, eliminating isolated noisy edges. The edge detection results are output as a binary image, where edge pixels are marked as 1 and non-edge pixels as 0. The set of edge point coordinates is stored in an array, formatted as (x, y) pixel positions, providing a preliminary data foundation for the crack contour.

[0035] Step S13: Perform contour repair based on crack edge data to obtain surface crack data; In this embodiment, contour repair is performed on the crack edge data output in step S12 to compensate for the discontinuity of crack edges caused by acquisition noise or breakpoints. Morphological closing operations (closing operations first dilate then erode) are used, with a 3×3 square kernel as the structuring element and two dilation iterations to connect edge breakpoints no more than 3 pixels apart. Subsequently, curve fitting is used to repair the crack contour gaps, specifically cubic B-spline interpolation to fit the missing contour segments, with an interpolation point spacing of 1 pixel. Geometric features are extracted from the repaired crack contour, including crack length (pixel count multiplied by the image scale bar of 0.1 mm / pixel), average width (measured along the contour normal direction using a grayscale threshold), and crack orientation angle (calculated using principal component analysis to determine the main direction of the crack point cloud). The repaired crack contour is stored as a polygonal point sequence and output as surface crack data.

[0036] Step S14: Identify network cracks based on surface crack data to obtain network crack data; In this embodiment, network cracks are identified based on the surface crack data obtained in step S13. The determination of network cracks is based on the number of crack branches and their angle characteristics. Topological structure analysis is performed on each crack profile to extract the locations of crack branch points, defined as points with a local node degree ≥ 3. The number of branch points for each crack is counted, with a threshold set to 3 or more. The angle between branches is calculated using the vector dot product method, with an angle threshold limited to branches within 45 degrees considered as network structures. A crack network graph is established using graph theory, where nodes are branch points and edges are crack branches. Crack groups meeting the conditions for the number of branches and the angle are marked as network cracks. The output network crack data includes branch point coordinates, crack skeleton line segment information, and corresponding topological relationships. The data is stored in an adjacency matrix format for easy subsequent analysis of abnormal crack growth.

[0037] Step S15: Identify transverse cracks based on surface crack data to obtain transverse crack data.

[0038] In this embodiment, transverse cracks are identified based on the surface crack data from step S13. First, a transverse crack is defined as a crack whose angle with the longitudinal direction of the dam (set as the vertical direction in the image coordinate system) is greater than 75 degrees and less than 105 degrees. The crack's main orientation angle (obtained using principal component analysis in step S13) is calculated, and crack profiles meeting the angle range are filtered. Second, the crack length must be greater than 500 mm, calculated based on the crack profile's pixel count multiplied by 0.1 mm / pixel. Cracks meeting the angle and length conditions are further validated for width, with a lower limit of 0.5 mm to exclude microcracks. Finally, cracks meeting the above conditions are marked as transverse cracks, and the output transverse crack data includes the crack's start and end point coordinates, length, width, and main orientation angle, used for subsequent crack width and width abrupt change anomaly analysis.

[0039] Preferably, step S14 specifically includes: Step S141: Extract crack centerline information based on surface crack data; In this embodiment, the surface crack polygonal contour data output in step S13 is used as input to extract the crack centerline. First, the polygonal contour is converted into a binary image, with the pixel value of the crack area set to 1 and the background set to 0. A thinning process based on a skeletonization algorithm is then employed, specifically the Zhang-Suen thinning algorithm, to iteratively erode the edge pixels of the crack area until a skeleton line with a single pixel width is obtained. The maximum number of iterations during the thinning process is set to 50, and the iteration terminates when the skeleton no longer changes. The skeleton obtained by this algorithm is the crack centerline, which is saved as a point set data, with the point spacing maintained at 0.1 mm / pixel, matching the image resolution. The skeleton nodes and their connections are stored in a linked list structure, providing backbone data for subsequent crack network analysis.

[0040] Step S142: Identify crack intersection relationships based on crack centerline information; In this embodiment, the crack centerline point set obtained in step S141 is used to identify crack intersection relationships. First, a topological analysis is performed on the crack skeleton to calculate the adjacency of all skeleton points. Adjacency is defined as the number of directly adjacent points connecting skeleton points. Intersection points are defined as points with an adjacency of ≥3, representing crack branching or intersection locations. Spatial clustering is performed on all intersection points, with a cluster radius of 0.3 mm to merge nearest neighbor intersection points caused by errors. The identified intersection point coordinates and their associated skeleton line segments constitute the crack intersection structure data. The intersection relationship data is stored in a graph structure, with nodes representing intersection points and edges representing skeleton line segments connecting them, and an adjacency matrix is ​​used for easy calculation.

[0041] Step S143: Calculate the crack connectivity based on the identified crack intersection relationships; In this embodiment, the crack connectivity is calculated based on the crack intersection diagram structure obtained in step S142. Crack connectivity is defined as the ratio of nodes (intersections) to edges (crack skeleton segments) in the crack network. The specific calculation formula is C=2E / (N(N-1)), where N is the number of intersections and E is the number of skeleton segments. First, the number of nodes N and edges E in the diagram are counted. The number of nodes depends on the number of intersections, and the number of edges is the total number of skeleton segments. To ensure calculation accuracy, short skeleton segments with a length less than 1 mm are removed. All valid edges are counted and traversed using an adjacency matrix. The crack connectivity result is between 0 and 1, reflecting the tightness of the crack network's connection.

[0042] Step S144: Predict the crack network trend based on crack connectivity; In this embodiment, a network trend judgment criterion is set based on the crack connectivity value calculated in step S143. If the connectivity C is greater than 0.4, the crack is determined to be in a state of obvious network trend. This threshold is obtained by statistically analyzing a large amount of historical crack image data to distinguish between ordinary cracks and network cracks. The prediction process combines connectivity with time series data to calculate the time derivative of connectivity ΔC / Δt, using a sliding window method with a window size of 24 hours and a step size of 1 hour. If the rate of change of connectivity is greater than 0.02 for three consecutive windows, the crack network trend is confirmed to have increased significantly. Based on this trend, the evolution of the crack into a network structure is predicted. The prediction result is stored as a Boolean variable, indicating whether the crack has entered the network stage.

[0043] Step S145: Mark the network crack region based on the crack networking trend; In this embodiment, based on the crack network trend predicted in step S144 and combined with the spatial distribution of crack centerlines, a network crack region is marked. First, the crack centerlines are divided into grid cells according to their spatial location, with a grid size of 5 mm × 5 mm. The number of crack connection points and the total crack length within each grid are counted. If the number of connection points is ≥3 and the total crack length is ≥10 mm, the grid is marked as a candidate network crack region. Further, based on the network trend judgment in step S144, if the crack connectivity degree within the region consistently exceeds 0.4 and the connectivity degree change rate consistently exceeds 0.02, the grid is finally marked as a network crack region. The marking results are stored as a grid index and corresponding status identifier for easy subsequent data analysis and display.

[0044] Step S146: Calculate the crack density based on the network crack region; perform high-density crack region clustering based on the crack density to obtain network crack data.

[0045] In this embodiment, crack density is calculated based on the network crack region marked in step S145. Crack density is defined as the total length of crack skeleton lines per unit area. The calculation method is to multiply the number of crack skeleton pixels within a statistical grid cell by the image resolution (0.1 mm / pixel), and then convert it to crack length in millimeters. The density threshold is set to be greater than 50 mm / cm². For grid cells exceeding this threshold, their spatial coordinates are extracted for density clustering analysis. The distance-based DBSCAN algorithm is used, with clustering parameters set to a neighborhood radius ε of 10 mm and a minimum sample size MinPts of 3, forming a spatially continuous high-density crack cluster region. The clustering results are output in the form of polygonal boundaries, which are formed by connecting all grid cells in the cluster, creating the final network crack dataset. This data includes cluster region coordinates, crack density statistics, and crack connectivity indices, serving as the basis for subsequent crack abnormal growth and fatigue analysis.

[0046] Preferably, step S15 specifically includes: Step S151: Calculate the crack orientation angle based on the surface crack data to obtain the crack orientation angle data; In this embodiment, based on the surface crack contour polygon data obtained in step S13, the crack orientation angle is calculated using the boundary line segment fitting method. The specific operations include: first, discretizing the crack contour into several line segments, each composed of the coordinates of adjacent vertices. For each line segment, its orientation vector is calculated, and the angle θ between the line segment and the horizontal direction is calculated using the arctangent function (atan2), with the angle unit being degrees and a value range of [0°, 180°]. To avoid the periodicity problem of the orientation angle, the angle is uniformly normalized to 0° to 180°. After calculation, the orientation angles of adjacent line segments are weighted and averaged according to their lengths, and short line segments with a length less than 1 mm are removed to eliminate noise interference. The final output is a dataset of the orientation angles of each skeleton line segment in the crack, stored as a structured table containing the starting coordinates, ending coordinates, length, and corresponding orientation angle of the line segment.

[0047] Step S152: Determine suspected horizontally distributed cracks based on crack orientation angle data, and obtain suspected horizontally distributed crack data; In this embodiment, based on the crack orientation angle data obtained in step S151, an angle threshold is used to screen and identify suspected horizontally distributed cracks. Specifically, the orientation angle threshold range is set to 85° to 95°, which corresponds to a near-horizontal (90°) direction. All crack segment orientation angle data are traversed, and segments whose orientation angles fall within this threshold range are marked as suspected horizontally distributed cracks. To avoid misjudgment of a single segment, and considering spatial continuity requirements: if the orientation angles of adjacent segments both meet the threshold conditions and the distance between their start and end points is less than 2 mm, they are merged into a continuous horizontal crack segment. Through this method, all suspected horizontal crack segments are filtered and summarized, outputting a suspected horizontally distributed crack dataset, including the start and end coordinates, length, orientation angle, and crack number of each crack segment.

[0048] Step S153: Detect stress concentration areas in the dam body based on suspected horizontally distributed crack data; In this embodiment, stress concentration areas are detected by combining suspected horizontal crack data with dam structural design parameters. First, a finite element mesh model of the dam structure is established, with the mesh size controlled within 10 cm to ensure local stress accuracy. Suspected horizontal crack segments are mapped onto the surface of the dam model, defining a crack influence radius of 5 cm, covering the finite element elements surrounding the crack. Actual dam loads, including water pressure, seismic force, and temperature stress, are applied to the finite element elements, and the stress distribution of the elements is calculated using elastoplastic mechanics. The criterion for determining a stress concentration area is that the maximum principal stress of the element exceeds 80% of the material's yield strength (the specific value depends on the material; for example, if the yield strength of concrete is set to 25 MPa, then the threshold is 20 MPa). Element regions that meet the threshold are marked as stress concentration areas, and the corresponding spatial coordinate range and mesh index are output.

[0049] Step S154: Conduct structural expansion detection based on the stress concentration area of ​​the dam body to obtain structural expansion data; In this embodiment, the stress concentration region marked in step S153 is used as the object for structural expansion detection. Multi-temporal optical image data and 3D point cloud data are used, and digital image correlation (DIC) technology is employed to measure the minute surface deformation of the stress concentration region. First, at least two sets of high-resolution images (resolution better than 0.05 mm / pixel) at different time points are acquired from the monitoring device, and the displacement field within the region is calculated using a sub-pixel-level correlation algorithm. By solving the displacement field gradient, the local surface expansion rate is calculated; the expansion rate is defined as the relative increase in area per unit area. An expansion rate threshold of 0.002 (i.e., 0.2%) is set; regions exceeding this threshold are recorded as structural expansion regions. The structural expansion data includes an expansion rate distribution matrix and corresponding spatial coordinates, and the storage format supports interfacing with finite element models.

[0050] Step S155: Determine the failure of steel reinforcement anchorage based on structural expansion data, and obtain steel reinforcement anchorage failure data; In this embodiment, the structural expansion data from step S154, combined with the reinforcement layout and anchorage design parameters, is used to determine the reinforcement anchorage failure status. Reinforcement anchorage failure is judged based on an expansion rate exceeding 0.0035 (0.35%) in the anchorage zone; this threshold is derived from laboratory reinforcement anchorage performance tests. The structural expansion rate data is combined with the reinforcement layout diagram, and the corresponding expansion rate value is extracted for each anchored reinforcement location. If the expansion rate of at least five consecutive pixels within the anchorage zone exceeds the threshold, the anchorage area is marked as failed. The spatial location and expansion rate statistics of the failed areas are summarized to form a reinforcement anchorage failure dataset. The data format includes the anchored reinforcement number, the coordinates of the start and end points of failure, and the failure severity index.

[0051] Step S156: Identify transverse cracks based on the rebar anchorage failure data to obtain transverse crack data.

[0052] In this embodiment, transverse cracks in the dam body are determined by combining the rebar anchorage failure data from step S155. A transverse crack is defined as a crack perpendicular or nearly perpendicular to the main load-bearing direction of the dam body (usually vertical). Using the surface crack data from step S13 and the spatial overlap with the anchorage failure area, the crack profile is filtered by direction angle. The filtering criteria are that the crack direction angle is between 0° and 15° and between 165° and 180° (representing an approximately vertical direction), and the spatial overlap area between the crack profile and the anchorage failure area exceeds 20%. Through spatial Boolean operations, crack profiles meeting the above conditions are extracted as transverse crack data. The output data includes the crack number, crack start and end coordinates, crack length, and corresponding anchorage failure block number, achieving quantitative spatial identification of transverse cracks.

[0053] Preferably, step S2 specifically includes: Step S21: Identify the crack width based on the transverse crack data to obtain the crack width data; In this embodiment, high-resolution digital image analysis technology is used to identify crack width using transverse crack contour data. First, grayscale enhancement and edge extraction are performed on the crack area image. The Canny edge detection algorithm is used, with parameters set to a low threshold of 30 and a high threshold of 90 to ensure the accuracy of crack edge detection. The extracted edge pixels constitute the crack boundary line. The crack width distribution is obtained by calculating the shortest distance between the inner and outer edge lines of the crack boundary. Specifically, for each point on the crack skeleton centerline, its vertical distance to the corresponding boundary line is calculated, and the distances to the left and right edges are recorded. The width is defined as the sum of the two distances. When converting pixel distance to actual width, it is set to 0.05 mm / pixel based on the known image spatial resolution parameters. The width data is stored in millimeters, and a crack width vector is output, containing the crack number, the coordinates of the crack centerline point, and the corresponding width value.

[0054] Step S22: Calculate the width gradient change value based on the crack width data; In this embodiment, the gradient change of the crack width is calculated based on the crack width data obtained in step S21. Specifically, a first-order difference calculation is performed on the crack width vector. The crack width sequence is defined as W={w1,w2,...,wn}, and the gradient change sequence G={g1,g2,...,gn-1} is calculated, where gi=|wi+1-wi| / d, and d is the distance between adjacent sampling points along the crack centerline in millimeters. The value of d is 0.1 millimeters to ensure detailed spatial resolution. The gradient calculation result reflects the rate of change of the crack width along the crack direction. To filter out high-frequency noise, median filtering is applied to the gradient sequence G, with a filtering window size of 5 points. The filtered gradient change data is used for subsequent anomaly detection. The gradient data storage format includes the crack number, the coordinates of the crack centerline starting point, the gradient value, and the corresponding position.

[0055] Step S23: Identify abrupt slope points based on the width gradient change value; In this embodiment, in the width gradient change sequence of step S22, abrupt slope points of crack width change are identified. The threshold for determining abrupt slope points is set to 0.15 mm / mm, meaning the rate of change of crack width between adjacent sampling points exceeds 0.15. Specifically, the gradient change sequence G is compared against the threshold, and all points with gi > 0.15 are marked as candidate abrupt slope points. To avoid misjudgment due to single-point noise, a neighborhood consistency check is further adopted. If the gradients of the three sampling points on both sides of a candidate point are all below 0.05, the point is retained; otherwise, it is discarded. The finally selected points are the abrupt slope points of crack width change, and the abrupt point dataset containing point coordinates, crack number, and gradient value is output.

[0056] Step S24: Extract width mutation anomaly data based on mutation slope points; In this embodiment, the abrupt change slope points selected in step S23 are used as the core, and width abrupt change anomalies are extracted by combining the original crack width data. Specifically, for each abrupt change slope point, a range of 0.5 mm is extended to both sides, and the maximum and minimum width values ​​within this range are counted. If the difference between the maximum and minimum widths exceeds 0.3 mm, the area is identified as a width abrupt change anomaly zone. This judgment threshold is based on typical standards for concrete crack width anomalies. The width abrupt change anomaly zones are stored with the start and end coordinates of the range, crack number, and abrupt width difference, forming a width abrupt change anomaly dataset for subsequent time series analysis.

[0057] Step S25: Calculate the width mutation timestamp based on the width mutation anomaly data; collect camera monitoring logs based on the width mutation timestamps; calculate the camera overheating status based on the camera monitoring logs to obtain camera overheating status data; In this embodiment, combining the width mutation anomaly data from step S24, the timestamps of the corresponding crack width mutations in the time series are statistically analyzed. The timestamps are recorded by the sensor synchronization clock of the crack monitoring system, accurate to the second. For each anomaly interval, the time point of its first occurrence is extracted as the width mutation timestamp. The system log of the crack monitoring camera is called via an interface to locate the camera's operating status for the corresponding time period based on the timestamps. The camera monitoring log includes temperature sensor readings, with a temperature sampling frequency of once per second. The camera overheating threshold is defined as 70 degrees Celsius; if the temperature exceeds the threshold for 5 consecutive seconds, it is recorded as an overheating event. The start and end time periods of the overheating event are recorded to generate camera overheating status data, including the time period, camera number, and temperature curve.

[0058] Step S26: Detect continuous data packet loss based on camera overheating status data to obtain continuous data packet loss data; perform communication link interruption analysis based on continuous data packet loss data to obtain communication link interruption data.

[0059] In this embodiment, based on the camera overheating time period marked in step S25, the monitoring system data transmission log for that time period is extracted. The data transmission log records the data packet reception time and loss details. Data packet loss is defined as no valid data received for more than 3 consecutive seconds. For continuous loss detection, a sliding time window analysis method is used, with a window size of 10 seconds and a step size of 1 second, to count the duration of continuous packet loss within the window. If the packet loss time exceeds 3 seconds, it is marked as continuous loss. All continuous loss events are summarized, and continuous data packet loss data is output, including start time, end time, duration, and affected camera number. Based on the continuous loss data, combined with the communication link topology and device status information, link interruption analysis is performed. Link interruption is defined as the overlap between a continuous data packet loss event and a physical link failure record, with a time difference not exceeding 5 seconds. The communication link interruption data includes the link start and end nodes, disconnection time period, disconnection duration, and affected area, supporting subsequent system maintenance and fault location.

[0060] Preferably, step S3 specifically includes: Step S31: Calculate the crack propagation rate based on the network crack data; perform crack anomalous growth analysis based on the crack propagation rate to obtain crack anomalous growth data; In this embodiment, using mesh crack image data acquired over multiple time periods, the crack contour is first precisely extracted. Image differencing technology is employed to calculate the crack length change ΔL (in millimeters) within a time interval Δt (in days) by comparing the crack contour edge coordinates at two different time points. The crack propagation rate V is defined as ΔL / Δt, in millimeters per day. The crack length change value in each crack mesh cell is extracted, and the spatially distributed crack propagation rate field is calculated. To remove environmental interference and image errors, a propagation rate threshold of 0.05 millimeters per day is set; propagation rates below this threshold are considered normal fluctuations and are not included in the abnormal growth range. Abnormal crack growth is identified by marking regions with propagation rates exceeding this threshold, and outputting abnormal crack growth data, including crack mesh number, propagation rate value, and corresponding timestamp.

[0061] Step S32: Identify the stress concentration region of the crack based on the abnormal crack growth data; In this embodiment, based on the abnormal crack growth region obtained in step S31, stress concentration regions are identified by combining crack mechanics analysis. Using finite element analysis (FEA) software, the abnormal crack propagation rate regions are mapped to structural elements by inputting the finite element model of the dam structure and the corresponding locations of abnormal crack growth in the network. The stress distribution around the crack region is calculated, focusing on the equivalent stress σ_eq. Using the Von Mises stress criterion, the stress concentration threshold is defined as 70% of the dam's design strength. Elements with an equivalent stress σ_eq exceeding this threshold are marked as stress concentration regions. The stress concentration region data is output, including element number, stress value, abnormal crack growth rate, and coordinate information. To ensure identification accuracy, the size of the finite element model elements is controlled within 5 cm, and the material parameters used are a concrete elastic modulus of 3.0 × 10^4 MPa and a Poisson's ratio of 0.2.

[0062] Step S33: Calculate the dynamic load based on the crack stress concentration region to obtain dynamic load data; calculate the cyclic stress amplitude based on the dynamic load data; perform crack propagation simulation based on the cyclic stress amplitude to obtain crack propagation data; In this embodiment, based on the stress concentration area determined in step S32, the dam's operating parameters are input, including water pressure (kPa), vibration frequency (Hz), and ambient temperature change (°C), and the corresponding dynamic load is calculated. The dynamic load is calculated using the dynamic load formula F=m·a, where m is the stressed mass and a is the acceleration, acquired by on-site sensors at a sampling frequency of 200Hz. The dynamic load distribution is mapped to the stress concentration area, and the local stress response is calculated. The cyclic stress amplitude Δσ is obtained by calculating the maximum and minimum stress difference, in MPa. Crack propagation simulation is based on Paris's rule, calculating the crack length a growth rate da / dN=C·(ΔK)^m, where ΔK is the stress intensity factor range, calculated from the cyclic stress amplitude and crack size, with C taken as 1.0×10^-12 and m as 3.0, all units in the International System of Units (SI). During simulation, crack propagation is calculated according to the number of cycles corresponding to the time step ΔN, outputting the relationship between crack propagation length and time, forming crack propagation data. The initial values ​​of crack size and material fracture toughness parameters are based on field test data input to ensure accurate simulation.

[0063] Step S34: Predict the material fatigue life based on crack propagation data to obtain material fatigue data; In this embodiment, the fatigue life of key dam components is calculated using the crack propagation length variation data with the number of cycles in step S33. The fatigue life calculation formula in fracture mechanics is adopted, defining the fatigue life as the number of cycles N_f corresponding to the crack propagation length reaching the critical fracture length a_c. The critical fracture length a_c is determined according to material structure design specifications; for concrete components, a_c is taken as 50 mm. The fatigue life N_f is calculated using the cumulative Paris rule. The actual service life, in years, is converted based on the relationship between the cumulative number of cycles and operating time. The material fatigue data includes the initial crack length, critical length, crack propagation curve, and corresponding fatigue life value. The data format includes component number, predicted life, crack size evolution process, and predicted time range, supporting subsequent monitoring and maintenance decisions.

[0064] Step S35: Conduct fracture detection of dam components based on material fatigue data to obtain fracture data of dam components.

[0065] In this embodiment, fracture detection is performed by combining the material fatigue life prediction results from step S34 with the actual operating time of the dam body. The fracture determination condition is: when the actual operating time of a component exceeds 95% of the predicted fatigue life, the component is marked as being in a fracture risk state. The component operating time is extracted from the monitoring data, in years, accurate to two decimal places. Fracture detection combines strain gauge and acoustic emission sensor data to monitor local strain changes and acoustic emission event frequency near the crack. When the strain change exceeds the baseline value of 0.005 (unitless strain) and the acoustic emission event frequency exceeds 10 times per minute, the fracture state is further confirmed. Fracture data includes component number, fracture risk level, strain value, acoustic emission event statistics, and timestamp. Component fracture data is used for system alarms and subsequent maintenance plan formulation. The data is stored in a structured format for easy retrieval and analysis.

[0066] Preferably, step S35 specifically includes: Step 351: Calculate the cumulative fatigue damage value based on material fatigue data; In this embodiment, based on the material fatigue life data obtained in the previous steps, the number of cycles for each part of the dam body under different cyclic load amplitudes is statistically analyzed. The cyclic load amplitude is divided into multiple intervals, such as 0 to 5 MPa, 5 to 10 MPa, and 10 to 15 MPa, with each interval corresponding to a fatigue life value. The actual number of cycles experienced within each amplitude interval is statistically analyzed using cyclic stress data collected by the monitoring system. The actual number of cycles is compared with the corresponding fatigue life, and the damage ratio of each interval is calculated. The damage ratios of all intervals are summed to obtain the total cumulative fatigue damage value. This cumulative value reflects the degree of fatigue experienced by the material, ranging from 0 to 1, where 0 represents no fatigue damage and 1 represents fatigue damage reaching its limit. The calculation process is performed in a dedicated data processing unit, with inputs including a cycle count statistics table and fatigue life data, and the output being a single scalar value.

[0067] Step 352: Identify highly fatigued areas of dam components based on cumulative fatigue damage values; In this embodiment, based on the cumulative fatigue damage value, the dam body is divided into several small grid cells, each corresponding to a cumulative fatigue damage value. All grid cells are traversed to identify cells with a cumulative fatigue damage value greater than or equal to a set threshold of 0.7; these cells are then classified as highly fatigued areas. To ensure accurate identification, cells with abnormal or missing monitoring data are excluded. The data for highly fatigued areas includes the spatial coordinates of the grid cells and their fatigue damage values, stored in a three-dimensional coordinate and numerical list format for convenient subsequent location and monitoring. The identification process is performed using dedicated software; the data input is a fatigue damage value matrix, and the output is a dataset of the spatial coordinates of the marked areas.

[0068] Step 353: Acquire acoustic emission signals based on the high fatigue region of dam components; perform signal preprocessing based on the acoustic emission signals to obtain the acoustic emission signals to be processed; calculate the peak amplitude based on the acoustic emission signals to be processed; calculate the signal energy based on the peak amplitude. In this embodiment, multiple acoustic emission sensors are arranged in the high fatigue region determined in step S352. Piezoelectric sensors with a frequency response range of 100 kHz to 1 MHz are used, and the sensors are fixed to the dam surface with a strong adhesive to ensure good coupling. The acquisition frequency is set to 2 MHz, and the acquisition time is 10 seconds. The acquired raw acoustic emission signal is preprocessed, including bandpass filtering to remove mechanical noise below 100 kHz and electronic interference above 900 kHz. Subsequently, a sliding time window method is used to divide the signal into multiple 1-millisecond segments, with a 0.5-millisecond overlap between segments. For the signal within each time segment, the maximum amplitude is measured as the peak amplitude, in volts. The energy is calculated based on the signal amplitude by summing the squares of the signal amplitudes within that time segment, with the result in volts square seconds. All peak amplitude and energy data are stored in chronological order for easy subsequent analysis.

[0069] Step 354: Determine the high-frequency acoustic emission signal energy region based on the signal energy; conduct strength degradation detection of dam component materials based on the high-frequency acoustic emission signal energy region to obtain dam component material degradation data; In this embodiment, based on the signal energy data calculated in step S353, an energy threshold of 1.0 × 10⁻⁶ volts square seconds is set. Any time segment with energy exceeding this threshold and its corresponding spatial region are identified as a high-energy acoustic emission signal region. The spatial location of this region is determined using triangulation based on the signal propagation delay between sensor arrays and known sensor coordinates. For these high-energy regions, the degree of material strength degradation is analyzed by combining material mechanical parameters such as elastic modulus, ultimate tensile strength, and fatigue characteristics obtained from prior experiments. Using calibration curves, the peak signal energy is correlated with the percentage decrease in material strength; typically, for every 1.0 × 10⁻⁷ volts square seconds increase in signal energy, the material strength decreases by 0.5%. The degradation data, including spatial coordinates, timestamps, and degradation percentages, is stored as a structured data table for subsequent structural evaluation.

[0070] Step 355: Based on the material degradation data of dam components, conduct fracture detection of dam components to obtain fracture data of dam components.

[0071] In this embodiment, based on the material degradation percentage data in step S354, it is determined that when the degradation level reaches or exceeds 50%, the component enters a fracture risk state. During fracture detection, strain gauge data from the dam body surface and crack monitoring data are used for auxiliary verification. Strain gauges are placed at key locations, with a sampling frequency of 1 Hz, monitoring whether the strain value exceeds the reference strain of 0.006. Crack monitoring uses ultrasonic testing and infrared thermal imaging technology to scan suspected fracture areas and confirm crack size and development trend. Fracture detection results record the component number, degradation level, strain anomaly value, crack length, and detection time. All data are stored in a structured format to support subsequent safety assessments and maintenance plan development.

[0072] Preferably, step S355 specifically includes: Calculate tensile strength based on the material degradation data of dam components; In this embodiment, using previously obtained data on the deterioration of dam components, including the percentage decrease in local material strength, damage location, and time, and combining it with the initial tensile strength benchmark value of the material (e.g., the tensile strength of ordinary concrete is typically 2.0 MPa), the current tensile strength of the material is calculated using a linear reduction method. Specifically, the initial tensile strength is multiplied by (1 minus the material deterioration percentage). For example, if the material deterioration rate is 0.3%, then the tensile strength is 2.0 MPa multiplied by 0.7, resulting in 1.4 MPa. This process utilizes the initial tensile strength parameters and deterioration percentage data stored in the database of the structural health monitoring system, and is automatically executed through a programming script. This ensures that each sampling point corresponds to a specific tensile strength value, and the output format is a spatial coordinate-tensile strength correspondence table for subsequent analysis.

[0073] The fatigue limit reduction factor is determined based on tensile strength. In this embodiment, the fatigue limit reduction factor is calculated based on the material's tensile strength data and the fatigue limit standard from laboratory fatigue test results. The fatigue limit reduction factor is defined as the ratio of the actual tensile strength to the material's standard tensile strength; this ratio reflects the degree to which the fatigue limit decreases due to material deterioration. Specifically, the tensile strength calculated at each spatial location is compared with the reference tensile strength. For example, if the reference tensile strength is 2.0 MPa and the actual tensile strength at a certain location is 1.5 MPa, then the reduction factor is 0.75. The calculation process uses the standard fatigue limit value recorded in the structural mechanics database (e.g., the fatigue limit of concrete is approximately 30% of its tensile strength). The reduction factor is multiplied by this standard value to obtain the current fatigue limit value. The data results are stored in the form of coordinate-fatigue limit reduction factor pairs for subsequent fatigue region calibration.

[0074] The fatigue zone of dam components is determined based on the fatigue limit reduction factor. In this embodiment, the spatial distribution of fatigue limit reduction coefficient data from all measuring points is analyzed, and a threshold-based determination method is used to identify fatigue regions. A fatigue limit reduction coefficient threshold of 0.8 is set; regions with reduction coefficients below this value are considered fatigue-sensitive regions. Specifically, the reduction coefficient data from the measuring points is interpolated and extended to the surface of the dam's 3D model to generate a reduction coefficient field distribution map. By partitioning and filtering the reduction coefficient field, continuous regions below 0.8 are extracted as fatigue regions. This process utilizes computer-aided design software to perform spatial data interpolation and region division, outputting the spatial coordinates and boundary shape of the fatigue regions, providing basic regional data for the construction of the three-dimensional model.

[0075] A three-dimensional model of dam components was constructed based on the fatigue region of dam components. In this embodiment, based on the fatigue region determined in the previous step, a three-modal model of the dam components is constructed, corresponding to the three principal vibration modes of the structure. Using finite element analysis software, a fine mesh is generated based on the geometric boundary data of the fatigue region, with the mesh element size controlled within 10 cm to ensure simulation accuracy. Mass, stiffness, and damping matrices are set for the three-modal analysis. The material parameters are the degraded elastic modulus and density, and the material elastic modulus reduction rate is adjusted according to the fatigue limit reduction factor. Eigenvalue analysis is used to solve for the three principal vibration frequencies and corresponding modal shapes. The output includes the three modal frequency values ​​and corresponding three-dimensional graphical data of the mode shapes. This model is used for subsequent high-temperature simulation and dynamic load response analysis.

[0076] High-temperature simulation was performed on the three-dimensional model of the dam components to obtain high-temperature data of the dam components; In this embodiment, based on a three-dimensional model and combined with historical data on actual temperature changes in the dam body (e.g., the highest daily temperature in summer can reach 60℃), a thermo-mechanical coupling simulation under high-temperature conditions is performed using the finite element thermal analysis method. Boundary conditions are applied to the temperature field, and the temperature distribution and thermal stress distribution inside the dam body are calculated using numerical solution techniques for the heat conduction equation. The coefficient of thermal expansion of the material is taken as 1.0×10^-5 / ℃, and the thermal conductivity coefficient is taken as 1.2W / (m·K) according to the material type. The simulation time step is set to 1 hour, and the simulation period is 30 days. The output is time-series temperature and thermal stress data, stored in a format including spatial coordinates and corresponding temperature values ​​and thermal stress vectors, which serve as input data for microscopic damage detection.

[0077] Microscopic damage data was obtained by detecting microscopic damage based on high-temperature data of dam components; In this embodiment, microscopic damage is detected based on thermal stress data obtained from high-temperature simulation and combined with material fatigue damage criteria. Using cumulative damage theory, the number of thermal stress cycles and stress amplitude for each grid cell are calculated. The thermal stress is compared with the material's yield strength (e.g., 20 MPa) to determine whether a damage state has been entered. By referencing the critical value for thermal fatigue crack initiation, a threshold is set where cells with a peak thermal stress exceeding 10 MPa and more than 500 consecutive cycles are marked as microscopic damage regions. The output microscopic damage data is a list of spatial coordinates and corresponding damage level values ​​(0-1 scale) for damage evolution analysis.

[0078] A thermally induced damage evolution map was drawn based on microscopic damage data; In this embodiment, microscopic damage data is used to create a thermally induced damage evolution map. The map employs 3D visualization technology, mapping spatial coordinates to a 3D point cloud. Damage levels are represented by color gradients, with low damage indicated in blue and high damage in red. Time-series microscopic damage data are superimposed sequentially to form a dynamic sequence of damage evolution. The map is generated using specialized graphics processing software, combined with time axis control, allowing users to observe the spatial distribution and trends of thermally induced damage within the dam body at different time points. The map data file contains spatial coordinates, timestamps, and corresponding damage level values.

[0079] The evolution of thermally induced cracks was analyzed based on the thermally induced damage evolution map to obtain thermally induced crack data; In this embodiment, crack initiation and propagation paths are identified based on a thermally induced damage evolution map. Continuous mesh elements with a damage level exceeding 0.7 are connected to form a crack network. The initial crack width is set to 0.1 mm, and as time progresses, the crack width increases linearly with the damage level, increasing by 0.05 mm per unit time. The crack propagation direction is determined by the direction of the maximum principal stress of thermal stress, and path simulation is performed using crack propagation rate data. The output thermally induced crack data includes crack spatial path coordinates, crack width, and propagation time series. The data format supports 3D cross-sectional display and dynamic demonstration.

[0080] Fracture prediction of dam components is performed based on thermally induced crack data, and fracture data of dam components are obtained.

[0081] In this embodiment, fracture risk is determined by combining thermally induced crack propagation data and fracture mechanics criteria. The fracture threshold is defined as a component entering a critical fracture state when the crack length reaches 100 mm and the width exceeds 0.5 mm. Fracture prediction is also based on a comprehensive assessment of crack propagation rate and cumulative fatigue damage data. Fracture data records the spatial location, time point, and degree of fracture occurrence, and is stored in a structured format for use in safety assessment and maintenance plan development.

[0082] Preferably, step S4 specifically includes: Step S41: Calculate the communication link interruption time based on the communication link interruption data; In this embodiment, communication link status data is extracted from the communication log database of the dam monitoring system. The data acquisition period is set to 1 second. The log records the link status of each monitoring node, marked as "connected" or "disconnected". A timestamp continuity detection method is used to identify time periods of continuous "disconnected" states. The duration of each link interruption is counted in seconds, with a duration threshold set to 5 seconds. Interruption events below this threshold are considered transient interference and are not included in the statistics. The statistical results include the start time, end time, and duration of the link interruption, stored in an event list format. This process automatically processes the log data using a programming language script, analyzing each log entry through loop traversal and conditional judgment to ensure accurate capture of all interruption events. The output interruption time data will be used to guide subsequent abnormal crack image acquisition, ensuring that the acquisition time point corresponds to the link interruption event.

[0083] Step S42: Acquire abnormal crack images based on the communication link interruption time; In this embodiment, the image acquisition time window of the crack monitoring camera is determined based on the link interruption time obtained in step S41. The image acquisition system is set to a 24-hour cyclic acquisition mode with a time resolution of once per minute. The abnormal crack image acquisition start condition is the time range from the start point to the end point of the link interruption time plus a buffer time of 1 minute before and after, ensuring that abnormal states occurring during and before the interruption are recorded. The camera is implemented using a high-resolution industrial camera with a resolution of 4096×2160 pixels and an exposure time fixed at 1 / 100 second to ensure image clarity. Image storage uses a lossless compression format (such as TIFF) and automatically adds timestamps and location identifiers. Image acquisition is automatically triggered by the field control unit, and corresponding image files are exported in batches according to the link interruption time period for subsequent crack analysis. All image files strictly correspond to the link interruption time to ensure spatiotemporal matching of data.

[0084] Step S43: Calculate the crack length based on the abnormal crack image; calculate the crack curvature based on the abnormal crack image; In this embodiment, crack length calculation employs an edge detection algorithm from image processing technology. Specifically, the Canny operator is used to preprocess the acquired abnormal crack images, with an upper threshold set to 150 and a lower threshold set to 50. After edge detection, morphological operations (erosion and dilation) are used to remove noise. Crack length is calculated by determining the maximum length of the connected region of the detected edge pixels. The conversion rate between pixels and actual distance is determined based on camera calibration data, specifically 0.05 mm / pixel. Curvature calculation is based on fitting a polynomial curve to the crack edge point sequence, using a third-order polynomial fitting method to calculate the curvature value of the curve at each point, in units of 1 / mm. The curvature threshold is set to 0.021 / mm to identify the degree of local bending of the crack. All image processing processes are executed automatically in batches, and the results are saved in the form of crack length (mm) and curvature distribution charts for easy subsequent morphological data integration.

[0085] Step S44: Integrate crack length and crack curvature to obtain crack morphology data; In this embodiment, the crack length and curvature data obtained in step S43 are integrated in a unified format to construct a crack morphology dataset. The crack morphology data includes key indicators such as total crack length, maximum curvature value, average curvature value, and curvature variation range. Specifically, the curvature distribution is segmented, distinguishing between low-curvature areas (curvature < 0.011 / mm), medium-curvature areas (0.01~0.021 / mm), and high-curvature areas (> 0.021 / mm), and the percentage of each segment relative to the total crack length is calculated. The integrated data is stored in a structured table, which includes fields such as crack ID, length (mm), maximum curvature (1 / mm), average curvature (1 / mm), high-curvature area length (mm), and high-curvature area percentage (%). This dataset serves as a standard description of crack morphology and is provided to the crack propagation direction prediction and display system, ensuring complete and structured data transmission.

[0086] Step S45: Predict the direction of crack propagation based on the fracture data of dam components; In this embodiment, the fracture data of dam components obtained in the aforementioned steps are used to extract the spatial initiation point of cracks and the normal vector information of the fracture surface. Fracture direction prediction is achieved by calculating the angle between the fracture surface normal vector and the local principal stress direction of the dam body. The principal stress directions are provided by the on-site stress monitoring system, specifically including three principal stress components and their spatial directions, with the unit being megapascals (MPa). During the prediction process, the angle between the fracture surface normal vector and the three principal stress directions is calculated. The crack propagation direction depends on the principal stress direction with the smallest angle, and the angle threshold is set to be within 15 degrees. The output crack propagation direction data is stored in vector form, containing the spatial initiation coordinates and three components (x, y, z) of the direction vector, for use by the crack display system.

[0087] Step S46: Transmit the crack morphology data and crack propagation direction to the crack display system to perform the abnormal crack display task.

[0088] In this embodiment, crack morphology data and crack propagation direction data are transmitted to the crack display system via a high-speed Ethernet interface. The transmission protocol is TCP / IP, and the data is encapsulated in JSON format to ensure cross-platform compatibility. Before transmission, data packets undergo CRC verification; packets failing the verification are retransmitted. After receiving the data, the crack display system uses a 3D visualization engine to spatially render the crack length, curvature distribution, and propagation direction, using different colors and arrows to identify the crack morphology and propagation trend. The display system synchronously displays historical and current crack states based on timestamps in the data, supporting interactive operations such as zooming and rotation. The entire data transmission and display process is executed automatically, with transmission latency controlled within 100 milliseconds, ensuring real-time visualization of crack anomaly information.

[0089] Preferably, this specification also provides an image recognition-based crack anomaly detection system for performing the image recognition-based crack anomaly detection method described above. The image recognition-based crack anomaly detection system includes: The crack type classification module is used to acquire monitoring images of hydropower station dams; perform surface crack analysis based on the monitoring images of hydropower station dams to obtain surface crack data; classify crack types based on surface crack data to obtain network crack data and transverse crack data; The communication link interruption analysis module is used to identify the crack width based on the transverse crack data to obtain crack width data; to perform width abrupt change anomaly detection based on the crack width data to obtain width abrupt change anomaly data; and to perform communication link interruption analysis based on the width abrupt change anomaly data to obtain communication link interruption data. The dam component fracture detection module is used to perform crack abnormal growth analysis based on network crack data to obtain crack abnormal growth data; perform material fatigue analysis based on crack abnormal growth data to obtain material fatigue data; and perform dam component fracture detection based on material fatigue data to obtain dam component fracture data. The crack display module is used to count the communication link interruption time based on the communication link interruption data; collect crack morphology based on the communication link interruption time to obtain crack morphology data; predict the crack propagation direction based on the fracture data of dam components; and transmit the crack morphology data and crack propagation direction to the crack display system to perform abnormal crack display tasks.

[0090] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0091] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A crack anomaly detection method based on image recognition, characterized in that, Includes the following steps: Step S1: Acquire monitoring images of the hydropower station dam; perform surface crack analysis based on the monitoring images of the hydropower station dam to obtain surface crack data; Crack types are classified based on surface crack data to obtain network crack data and transverse crack data; Step S2: Identify the crack width based on the transverse crack data to obtain crack width data; perform width abrupt change anomaly detection based on the crack width data to obtain width abrupt change anomaly data; Communication link interruption data is obtained by analyzing the width mutation anomaly data. Step S3: Perform crack anomalous growth analysis based on the network crack data to obtain crack anomalous growth data; Material fatigue data is obtained by performing material fatigue analysis based on crack abnormal growth data; Fracture detection of dam components is performed based on material fatigue data to obtain fracture data of dam components. Step S4: Calculate the communication link interruption time based on the communication link interruption data; collect crack morphology data based on the communication link interruption time; predict the crack propagation direction based on the dam component fracture data; transmit the crack morphology data and crack propagation direction to the crack display system to perform the abnormal crack display task.

2. The crack anomaly detection method based on image recognition according to claim 1, characterized in that, Step S1 is as follows: Step S11: Acquire monitoring images of the hydropower station dam; Step S12: Detect crack edges based on monitoring images of the hydropower station dam to obtain crack edge data; Step S13: Perform contour repair based on crack edge data to obtain surface crack data; Step S14: Identify network cracks based on surface crack data to obtain network crack data; Step S15: Identify transverse cracks based on surface crack data to obtain transverse crack data.

3. The crack anomaly detection method based on image recognition according to claim 2, characterized in that, Step S14 is as follows: Step S141: Extract crack centerline information based on surface crack data; Step S142: Identify crack intersection relationships based on crack centerline information; Step S143: Calculate the crack connectivity based on the identified crack intersection relationships; Step S144: Predict the crack network trend based on crack connectivity; Step S145: Mark the network crack region based on the crack networking trend; Step S146: Calculate the crack density based on the network crack region; High-density crack regions are clustered based on crack density to obtain network crack data.

4. The crack anomaly detection method based on image recognition according to claim 2, characterized in that, Step S15 is as follows: Step S151: Calculate the crack orientation angle based on the surface crack data to obtain the crack orientation angle data; Step S152: Determine suspected horizontally distributed cracks based on crack orientation angle data, and obtain suspected horizontally distributed crack data; Step S153: Detect stress concentration areas in the dam body based on suspected horizontally distributed crack data; Step S154: Conduct structural expansion detection based on the stress concentration area of ​​the dam body to obtain structural expansion data; Step S155: Determine the failure of steel reinforcement anchorage based on structural expansion data, and obtain steel reinforcement anchorage failure data; Step S156: Identify transverse cracks based on the rebar anchorage failure data to obtain transverse crack data.

5. The crack anomaly detection method based on image recognition according to claim 1, characterized in that, Step S2 is as follows: Step S21: Identify the crack width based on the transverse crack data to obtain the crack width data; Step S22: Calculate the width gradient change value based on the crack width data; Step S23: Identify abrupt slope points based on the width gradient change value; Step S24: Extract width mutation anomaly data based on mutation slope points; Step S25: Calculate the width mutation timestamp based on the width mutation anomaly data; Camera monitoring logs are collected based on width-increment timestamps. The camera overheating status data is obtained by analyzing the camera monitoring logs. Step S26: Detect continuous data packet loss based on camera overheating status data to obtain continuous data packet loss data; Communication link interruption data is obtained by analyzing the continuous loss of data packets.

6. The crack anomaly detection method based on image recognition according to claim 1, characterized in that, Step S3 is as follows: Step S31: Calculate the crack propagation rate based on the network crack data; perform crack anomalous growth analysis based on the crack propagation rate to obtain crack anomalous growth data; Step S32: Identify the stress concentration region of the crack based on the abnormal crack growth data; Step S33: Calculate the dynamic load based on the stress concentration region of the crack to obtain the dynamic load data; Calculate the cyclic stress amplitude based on dynamic load data; Crack propagation simulation was performed based on the cyclic stress amplitude to obtain crack propagation data; Step S34: Predict the material fatigue life based on crack propagation data to obtain material fatigue data; Step S35: Conduct fracture detection of dam components based on material fatigue data to obtain fracture data of dam components.

7. The crack anomaly detection method based on image recognition according to claim 6, characterized in that, Step S35 is as follows: Step 351: Calculate the cumulative fatigue damage value based on material fatigue data; Step 352: Identify highly fatigued areas of dam components based on cumulative fatigue damage values; Step 353: Acquire acoustic emission signals based on the high fatigue region of dam components; perform signal preprocessing based on the acoustic emission signals to obtain the acoustic emission signals to be processed; calculate the peak amplitude based on the acoustic emission signals to be processed; calculate the signal energy based on the peak amplitude. Step 354: Determine the high-frequency acoustic signal energy region based on the signal energy; Based on the energy region of high-frequency acoustic emission signals, the strength degradation detection of dam component materials is carried out to obtain dam component material degradation data; Step 355: Based on the material degradation data of dam components, conduct fracture detection of dam components to obtain fracture data of dam components.

8. The crack anomaly detection method based on image recognition according to claim 7, characterized in that, Step S355 specifically includes: Calculate tensile strength based on the material degradation data of dam components; The fatigue limit reduction factor is determined based on tensile strength. The fatigue zone of dam components is determined based on the fatigue limit reduction factor. A three-dimensional model of dam components was constructed based on the fatigue region of dam components. High-temperature simulation was performed on the three-dimensional model of the dam components to obtain high-temperature data of the dam components; Microscopic damage data was obtained by detecting microscopic damage based on high-temperature data of dam components; A thermally induced damage evolution map was drawn based on microscopic damage data; The evolution of thermally induced cracks was analyzed based on the thermally induced damage evolution map to obtain thermally induced crack data; Fracture prediction of dam components is performed based on thermally induced crack data, and fracture data of dam components are obtained.

9. The crack anomaly detection method based on image recognition according to claim 1, characterized in that, Step S4 is as follows: Step S41: Calculate the communication link interruption time based on the communication link interruption data; Step S42: Acquire abnormal crack images based on the communication link interruption time; Step S43: Calculate the crack length based on the abnormal crack image; Calculate crack curvature based on abnormal crack images; Step S44: Integrate crack length and crack curvature to obtain crack morphology data; Step S45: Predict the direction of crack propagation based on the fracture data of dam components; Step S46: Transmit the crack morphology data and crack propagation direction to the crack display system to perform the abnormal crack display task.

10. A crack anomaly detection system based on image recognition, characterized in that, For performing the image recognition-based crack anomaly detection method as described in claim 1, the image recognition-based crack anomaly detection system comprises: The crack type classification module is used to acquire monitoring images of hydropower station dams; perform surface crack analysis based on the monitoring images of hydropower station dams to obtain surface crack data; classify crack types based on surface crack data to obtain network crack data and transverse crack data; The communication link interruption analysis module is used to identify the crack width based on the transverse crack data to obtain crack width data; to perform width abrupt change anomaly detection based on the crack width data to obtain width abrupt change anomaly data; and to perform communication link interruption analysis based on the width abrupt change anomaly data to obtain communication link interruption data. The dam component fracture detection module is used to perform crack abnormal growth analysis based on network crack data to obtain crack abnormal growth data; perform material fatigue analysis based on crack abnormal growth data to obtain material fatigue data; and perform dam component fracture detection based on material fatigue data to obtain dam component fracture data. The crack display module is used to count the communication link interruption time based on the communication link interruption data; collect crack morphology based on the communication link interruption time to obtain crack morphology data; predict the crack propagation direction based on the fracture data of dam components; and transmit the crack morphology data and crack propagation direction to the crack display system to perform abnormal crack display tasks.

Images