Intelligent perception and early warning system for full-life damage of hybrid tower structure based on digital twinning

Abstract

The application discloses a full-life damage intelligent perception and early warning system for a hybrid tower structure based on digital twinning, relates to the technical field of hybrid tower damage monitoring and early warning, and comprises an inspection image acquisition module, which performs time label correlation processing, spatial position correlation processing and part corresponding arrangement processing on tower foundation area images, concrete tower section images, steel and concrete connecting section images, upper tower section images and shooting pose information, and generates hybrid tower inspection image data.In the application, the image content and shooting pose information of different inspection time points, different shooting angles and different hybrid tower structure parts are uniformly correlated, the problems of dispersed inspection images, inconsistent angles, discontinuous part connection and difficult unified correspondence of surface positions are solved, the full-surface image content of the hybrid tower structure can be continuously organized, uniformly expressed and stably mapped to a three-dimensional surface, and the integrity, spatial consistency and digital twinning reconstruction accuracy of the hybrid tower structure inspection result are improved.

Description

Technical Field

[0001] This invention relates to the field of mixed tower damage monitoring and early warning technology, and in particular to a smart sensing and early warning system for the whole life cycle damage of mixed tower structures based on digital twins. Background Technology

[0002] Hybrid tower structures combine multiple structural components, including concrete tower sections, steel-concrete connecting sections, and upper tower sections. They have long service cycles and are subject to continuous environmental loads, thus placing high demands on inspection and monitoring, damage identification, and operational status assessment. With the development of digital operation and maintenance and intelligent structural monitoring, combining inspection images, three-dimensional surface models, and time-series analysis results has become an important development direction for the whole life cycle monitoring of hybrid tower structures. This also provides a technical background for further analysis of the shortcomings of existing technologies in continuous damage perception at multiple time points, from multiple perspectives, and at multiple locations.

[0003] Existing technologies for inspecting and monitoring damage in mixed tower structures primarily rely on the separate acquisition and processing of images from different inspection points, shooting angles, and tower parts. The lack of a unified correlation between image content, shooting pose, time information, and structural parts easily leads to scattered inspection data, inconsistent perspectives, and discontinuous connections between parts. Consequently, it becomes difficult to establish a stable correspondence between multi-view images and the same structural surface. Current methods generally lack a continuous comparison mechanism combined with digital twin surfaces, making it difficult to stably track damage boundaries, coverage areas, and expansion directions. This results in unclear damage evolution processes, unintuitive changes before and after damage, and difficulty in establishing a unified correspondence between early warning information and specific tower parts, affecting the continuity and accuracy of damage perception and early warning throughout the entire lifespan of mixed tower structures. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a digital twin-based intelligent perception and early warning system for the full life-cycle damage of hybrid tower structures.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a digital twin-based intelligent perception and early warning system for the full life-cycle damage of hybrid tower structures, comprising: an inspection image acquisition module, which performs time-stamping association processing, spatial location association processing, and part correspondence processing on images of the tower base area, concrete tower sections, steel-concrete connection sections, and upper tower sections, as well as the shooting pose information, to generate hybrid tower inspection image data; a perspective unification processing module, which performs image distortion correction processing, image scale unification processing, image perspective registration processing, and tower surface unfolding association processing on the hybrid tower inspection image data, to generate perspective unified image data; a 3D twin modeling module, which performs multi-view reconstruction processing, surface point cloud generation processing, tower surface mesh fitting processing, and texture mapping processing on the perspective unified image data, to generate hybrid tower 3D surface model data, and generates hybrid tower digital twin data based on the hybrid tower 3D surface model data; and a damage image analysis module, The system performs crack region segmentation, spalling region segmentation, corrosion region segmentation, and damage edge contour refinement on the unified viewpoint image data to generate damage segmentation result data. The damage 3D projection module performs damage region pose projection processing, surface impact point correspondence processing, homogeneous damage spatial merging processing, and duplicate mapping reduction processing on the damage segmentation result data and the mixed tower 3D surface model data to generate damage 3D projection data. The temporal twin mapping module performs time alignment processing, spatial position inheritance processing, damage region expansion comparison processing, and twin surface evolution writing processing on the damage 3D projection data corresponding to different inspection time points to generate damage temporal mapping data. The early warning output module performs regional change rate analysis processing, continuous change analysis processing of tower body parts, damage expansion direction analysis processing, and early warning result generation processing on the damage temporal mapping data to output damage evolution early warning data and mixed tower digital twin visualization early warning results.

[0006] As a further description of the above technical solution: The inspection image acquisition module receives images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, as well as the corresponding shooting pose information, collected at different inspection times for the hybrid tower structure. It performs time-stamping association processing on the tower base area, concrete tower section, steel-concrete connection section, and upper tower section images according to the inspection time. It performs spatial location association processing on the shooting pose information and corresponding images. It performs part-to-part correspondence processing on the image content and hybrid tower structure parts. Finally, it performs unified integration processing on the results of the time-stamping association processing, spatial location association processing, and part-to-part correspondence processing to generate hybrid tower inspection image data.

[0007] As a further description of the above technical solution: The unified view processing module receives mixed tower inspection image data and extracts images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, as well as shooting pose information, from the mixed tower inspection image data. It performs image distortion correction and image scale unification processing on the images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section. Combining the shooting pose information, it performs image view registration processing on the images after image distortion correction and image scale unification processing. It performs tower surface expansion association processing on the images after image view registration processing. Finally, it performs association writing processing on the results after tower surface expansion association processing to generate unified view image data.

[0008] As a further description of the above technical solution: The 3D twin modeling module receives unified view image data and extracts image location content, image texture content, and surface unfolding location content from it. It then performs multi-view reconstruction processing on these data to generate corresponding reconstruction results. Next, it performs surface point cloud generation and tower surface mesh fitting processing on the reconstructed results. Finally, it performs texture mapping processing on the tower surface mesh fitting results and the image texture content to generate hybrid tower 3D surface model data. Finally, it performs twin surface correspondence writing processing on the hybrid tower 3D surface model data to generate hybrid tower digital twin data.

[0009] As a further description of the above technical solution: The damage image analysis module receives unified view image data and mixed tower 3D surface model data. It extracts initial damage seed regions from the crack, spalling, and corrosion areas, as well as adjacent view images, from the unified view image data. Using a seed region growth algorithm, and combining the boundary correspondences in adjacent view images, the corresponding surface positions in the mixed tower 3D surface model data, and the corresponding location relationships of the tower base, concrete tower section, steel-concrete connection section, and upper tower section, it performs seed preservation processing on each initial damage seed region, forming a preserved damage seed region within the corresponding location range. The preserved damage seed regions then undergo boundary expansion processing to maintain consistency with the current damage seed region in terms of grayscale variation, texture direction, boundary continuity, and location continuity. Adjacent pixels are merged into the current damage seed region, and adjacent pixels that are discontinuous with the current damage seed region in terms of texture direction deviation and corresponding surface position offset are excluded from the boundary expansion range, forming the expanded damage region boundary. Feedback write-back processing is performed on the expanded damage region boundary. Based on the alignment results between the expanded damage region boundary and the corresponding boundary in the adjacent view image, the fitting results between the expanded damage region boundary and the corresponding surface position in the mixed tower 3D surface model data, and the continuity results between the boundary boundary and the corresponding part position relationship, continuous and stable boundary directions are further expanded, and boundary retraction is performed on continuously offset boundary directions with discontinuous part position relationships. After completing the further expansion and boundary retraction, the damage region boundary is updated to form the damage segmentation result data.

[0010] As a further description of the above technical solution: The 3D damage projection module receives damage segmentation results and mixed tower 3D surface model data. It extracts damage boundary segments from images at different viewpoints from the damage segmentation results. A seed region growing algorithm is used to perform pose projection processing on the damage segmentation results and mixed tower 3D surface model data. Based on the surface position relationships of the tower base area, concrete tower segment, steel-concrete connection segment, and upper tower segment, damage boundary segments projected to the same location and with adjacent projection positions are pruned and correspondence-based. Correspondences inconsistent with the current projection position, boundary orientation, and location relationship are removed, resulting in pruned and correspondence-based damage boundary segments. The module then compares the pruned and correspondence-based damage boundary segments with the mixed tower 3D surface model data. The surface segments undergo point-to-surface convergence processing. Based on the fitting results of each damage boundary segment to the local surface normal direction, the positional continuity between local surface segments, and the surface connection relationship within the corresponding area, the projection position of each damage boundary segment in the hybrid tower 3D surface model data is updated to form updated damage boundary segments. The updated damage boundary segments undergo homogeneous damage spatial merging processing. Damage boundary segments that are continuous and consistent in spatial position, boundary orientation, surface fitting results, and location positional relationship are merged into the same damage region. The merged damage region undergoes repeated mapping reduction processing, and a single projection region corresponding to the damage region is retained in the hybrid tower 3D surface model data to form damage 3D projection data.

[0011] As a further description of the above technical solution: The temporal twin mapping module receives damage 3D projection data and mixed tower digital twin data; it performs time alignment processing on the damage 3D projection data corresponding to different inspection time points according to the inspection time point, generating time alignment results; it performs corresponding processing on the time alignment results and mixed tower digital twin data according to their spatial position relationship, generating spatial inheritance results; it uses a dynamic time warping algorithm combined with a Hampel filtering algorithm to perform damage region expansion comparison processing on the spatial inheritance results, time alignment results, and mixed tower digital twin data, generating expansion comparison results; and it performs twin surface evolution writing processing on the expansion comparison results to generate damage temporal mapping data.

[0012] As a further description of the above technical solution: The dynamic time warping algorithm performs damage region expansion comparison processing on the spatial inheritance results, time alignment results, and mixed tower digital twin data. The steps include: performing continuous region correspondence processing on the three-dimensional projection data of damage corresponding to the previous inspection time point and the current inspection time point in the spatial inheritance results according to their spatial position; and extracting the damage boundary, damage coverage area, and damage expansion direction from the previous inspection time point and the current inspection time point, based on the time alignment results, to form the original damage change sequence corresponding to the previous inspection time point and the original damage change sequence corresponding to the current inspection time point; and using the Hampel filtering algorithm to compare the damage region expansion of the previous inspection time point and the current inspection time point. The original damage change sequence corresponding to the previous inspection point and the original damage change sequence corresponding to the current inspection point are processed by sliding window anomaly identification and anomaly replacement, respectively. Segmentation and aggregation processing is then performed on continuous anomaly locations to form the smooth damage change sequence corresponding to the previous inspection point, the smooth damage change sequence corresponding to the current inspection point, the anomaly segmentation result corresponding to the previous inspection point, and the anomaly segmentation result corresponding to the current inspection point. Local distance calculation is then performed on the smooth damage change sequences corresponding to the previous and current inspection points to generate a base distance matrix. Finally, the base distance matrix is ​​anomaly-adjusted based on the anomaly segmentation results corresponding to the previous and current inspection points. A constant position correction process is performed to generate a corrected distance matrix. This corrected distance matrix is ​​then combined with the abnormal segment results from the previous and current inspection points to perform a regularization window shrinkage process. The regularization window width is reduced for continuously changing segments, while a local elastic regularization window is retained for segments corresponding to abnormal segments, generating a regularized window result. Within this regularized window result, a step cost renormalization process is performed on the corrected distance matrix. The cumulative cost is increased for matching units within abnormal segments and those crossing abnormal segment boundaries, while the base cost is maintained for matching units within continuously changing segments, generating a step renormalization result. Finally, a forward regularization path search and a reverse regularization path search process are performed on the step renormalization result. The system generates forward regularization path results and reverse regularization path results. It then performs bidirectional path consistency verification on these results, identifying segments with consistent or adjacent path mappings as stable alignment segments and segments with deviated path mappings and corresponding abnormal segments as unstable alignment segments, generating path consistency results. Finally, it performs write-back and rollback processing on the unstable alignment segments in the path consistency results, combining the abnormal segment results from the previous inspection time point, the abnormal segment results from the current inspection time point, and the corresponding surface areas in the mixed tower digital twin data. Within the rolled-back local regularization window, it re-executes dynamic time regularization path search processing to generate write-back and re-regularization results.A unified integration process is performed on the stable alignment sections and the write-back re-normalization results. This is combined with the corresponding surface regions in the hybrid tower digital twin data to perform boundary extension correspondence processing, coverage area correspondence processing, expansion direction correspondence processing, and position consistency verification processing, generating extended comparison results.

[0013] As a further description of the above technical solution: The early warning output module receives damage time-series mapping data and extracts regional change content, tower part change content, and damage expansion content from the damage time-series mapping data. It performs regional change rate analysis processing on the regional change content to generate regional change rate results; it performs continuous change analysis processing on the tower part change content to generate continuous change results; it performs damage expansion direction analysis processing on the damage expansion content to generate damage expansion direction results; and it performs early warning result generation processing on the regional change rate results, continuous change results of tower parts, and damage expansion direction results, outputting damage evolution early warning data and hybrid tower digital twin visualization early warning results.

[0014] The present invention has the following beneficial effects: 1. In this invention, the image content and shooting pose information of different inspection time points, different shooting angles, and different parts of the mixed tower structure are first uniformly associated, and then the image distortion correction, scale unification, perspective registration, surface unfolding, multi-view reconstruction, surface point cloud generation, tower surface mesh fitting and texture mapping are completed. This solves the problems of scattered inspection images, inconsistent perspectives, discontinuous parts, and difficulty in uniformly corresponding surface positions. It enables the full surface image content of the mixed tower structure to be continuously organized, uniformly expressed and stably mapped to the three-dimensional surface, improving the integrity, spatial consistency and digital twin reconstruction accuracy of the mixed tower structure inspection results.

[0015] 2. In this invention, by refining the damage edge contours of crack areas, spalling areas, and corrosion areas, and combining the corresponding surface positions, location relationships, damage area pose projection, surface landing point correspondence, spatial merging of homogeneous damage, and reduction of repeated mapping, the problems of two-dimensional damage boundaries being difficult to accurately land on the real structural surface, the same damage area being easily mapped repeatedly from different perspectives, and local boundaries being easily misaligned and dispersed are solved. This enables the damage area to achieve single-area landing, continuous boundary preservation, and stable spatial merging in the mixed tower three-dimensional surface model data, improving the positioning accuracy, location differentiation ability, and three-dimensional expression consistency of the damage identification results.

[0016] 3. In this invention, by performing time alignment, spatial position inheritance, damage area expansion comparison, and twin surface evolution writing on the three-dimensional projection data of damage corresponding to different inspection points, and further performing regional change rate analysis, continuous change analysis of tower body parts, damage expansion direction analysis, and early warning result generation, the problems of difficulty in continuously comparing the results of previous and subsequent inspections, difficulty in stably tracking the damage evolution process, and difficulty in uniformly associating early warning information with the spatial position of the structure are solved. This enables the damage boundary, damage coverage area, and damage expansion direction to be continuously inherited, dynamically depicted, and intuitively warned in both time and space dimensions, improving the continuity of damage perception throughout the entire life cycle of the hybrid tower structure, the pertinence of early warning judgment, and the intuitiveness of the visualized early warning results. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Reference Figure 1This invention provides an embodiment of a digital twin-based intelligent perception and early warning system for the full life-cycle damage of hybrid tower structures, comprising: an inspection image acquisition module, which performs time-stamped association processing, spatial location association processing, and part correspondence processing on images of the tower base area, concrete tower sections, steel-concrete connection sections, and upper tower sections, along with the shooting pose information, to generate hybrid tower inspection image data; a perspective unification processing module, which performs image distortion correction processing, image scale unification processing, image perspective registration processing, and tower surface unfolding association processing on the hybrid tower inspection image data, to generate perspective unification image data; a 3D twin modeling module, which performs multi-view reconstruction processing, surface point cloud generation processing, tower surface mesh fitting processing, and texture mapping processing on the perspective unification image data, to generate hybrid tower 3D surface model data, and generates hybrid tower digital twin data based on the hybrid tower 3D surface model data; and a damage image analysis module, which performs perspective unification... The image data undergoes crack region segmentation, spalling region segmentation, corrosion region segmentation, and damage edge contour refinement to generate damage segmentation result data. The damage 3D projection module performs damage region pose projection processing, surface impact point correspondence processing, homogeneous damage spatial merging processing, and duplicate mapping reduction processing on the damage segmentation result data and the mixed tower 3D surface model data to generate damage 3D projection data. The temporal twin mapping module performs time alignment processing, spatial position inheritance processing, damage region expansion comparison processing, and twin surface evolution writing processing on the damage 3D projection data corresponding to different inspection time points to generate damage temporal mapping data. The early warning output module performs regional change rate analysis processing, continuous change analysis processing of tower body parts, damage expansion direction analysis processing, and early warning result generation processing on the damage temporal mapping data, outputting damage evolution early warning data and mixed tower digital twin visualization early warning results.

[0020] The inspection image acquisition module receives images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, as well as the corresponding shooting pose information, collected at different inspection times for the hybrid tower structure. It performs time-stamping association processing on the tower base area, concrete tower section, steel-concrete connection section, and upper tower section images according to the inspection time. It performs spatial location association processing on the shooting pose information and corresponding images. It performs part-to-part correspondence processing on the image content and hybrid tower structure parts. Finally, it performs unified integration processing on the results of the time-stamping association processing, spatial location association processing, and part-to-part correspondence processing to generate hybrid tower inspection image data.

[0021] The unified view processing module receives mixed tower inspection image data and extracts images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, as well as shooting pose information, from the mixed tower inspection image data. It performs image distortion correction and image scale unification processing on the images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section. Combining the shooting pose information, it performs image view registration processing on the images after image distortion correction and image scale unification processing. It performs tower surface expansion association processing on the images after image view registration processing. Finally, it performs association writing processing on the results after tower surface expansion association processing to generate unified view image data.

[0022] The 3D twin modeling module receives unified view image data and extracts image location content, image texture content, and surface unfolding location content from it. It then performs multi-view reconstruction processing on these data to generate corresponding reconstruction results. Next, it performs surface point cloud generation and tower surface mesh fitting processing on the reconstructed results. Finally, it performs texture mapping processing on the tower surface mesh fitting results and the image texture content to generate hybrid tower 3D surface model data. Finally, it performs twin surface correspondence writing processing on the hybrid tower 3D surface model data to generate hybrid tower digital twin data.

[0023] The damage image analysis module receives unified view image data and mixed tower 3D surface model data. It extracts initial damage seed regions from the crack, spalling, and corrosion areas, as well as adjacent view images, from the unified view image data. Using a seed region growth algorithm, and combining the boundary correspondences in adjacent view images, the corresponding surface positions in the mixed tower 3D surface model data, and the corresponding location relationships of the tower base, concrete tower section, steel-concrete connection section, and upper tower section, it performs seed preservation processing on each initial damage seed region, forming a preserved damage seed region within the corresponding location range. The preserved damage seed regions then undergo boundary expansion processing to maintain consistency with the current damage seed region in terms of grayscale variation, texture direction, boundary continuity, and location continuity. Adjacent pixels are merged into the current damage seed region, and adjacent pixels that are discontinuous with the current damage seed region in terms of texture direction deviation and corresponding surface position offset are excluded from the boundary expansion range, forming the expanded damage region boundary. Feedback write-back processing is performed on the expanded damage region boundary. Based on the alignment results between the expanded damage region boundary and the corresponding boundary in the adjacent view image, the fitting results between the expanded damage region boundary and the corresponding surface position in the mixed tower 3D surface model data, and the continuity results between the boundary boundary and the corresponding part position relationship, continuous and stable boundary directions are further expanded, and boundary retraction is performed on continuously offset boundary directions with discontinuous part position relationships. After completing the further expansion and boundary retraction, the damage region boundary is updated to form the damage segmentation result data.

[0024] The 3D damage projection module receives damage segmentation results and mixed tower 3D surface model data. It extracts damage boundary segments from images at different viewpoints from the damage segmentation results. A seed region growing algorithm is used to perform pose projection processing on the damage segmentation results and mixed tower 3D surface model data. Based on the surface position relationships of the tower base area, concrete tower segment, steel-concrete connection segment, and upper tower segment, damage boundary segments projected to the same location and with adjacent projection positions are pruned and correspondence-based. Correspondences inconsistent with the current projection position, boundary orientation, and location relationship are removed, resulting in pruned and correspondence-based damage boundary segments. The module then compares the pruned and correspondence-based damage boundary segments with the mixed tower 3D surface model data. The surface segments undergo point-to-surface convergence processing. Based on the fitting results of each damage boundary segment to the local surface normal direction, the positional continuity between local surface segments, and the surface connection relationship within the corresponding area, the projection position of each damage boundary segment in the hybrid tower 3D surface model data is updated to form updated damage boundary segments. The updated damage boundary segments undergo homogeneous damage spatial merging processing. Damage boundary segments that are continuous and consistent in spatial position, boundary orientation, surface fitting results, and location positional relationship are merged into the same damage region. The merged damage region undergoes repeated mapping reduction processing, and a single projection region corresponding to the damage region is retained in the hybrid tower 3D surface model data to form damage 3D projection data.

[0025] The temporal twin mapping module receives damage 3D projection data and mixed tower digital twin data; it performs time alignment processing on the damage 3D projection data corresponding to different inspection time points according to the inspection time point, generating time alignment results; it performs corresponding processing on the time alignment results and mixed tower digital twin data according to their spatial position relationship, generating spatial inheritance results; it uses a dynamic time warping algorithm combined with a Hampel filtering algorithm to perform damage region expansion comparison processing on the spatial inheritance results, time alignment results, and mixed tower digital twin data, generating expansion comparison results; and it performs twin surface evolution writing processing on the expansion comparison results to generate damage temporal mapping data.

[0026] The dynamic time warping algorithm performs damage region expansion comparison processing on the spatial inheritance results, time alignment results, and mixed tower digital twin data. The steps include: performing continuous region correspondence processing on the three-dimensional projection data of damage corresponding to the previous inspection time point and the current inspection time point in the spatial inheritance results according to their spatial position; and extracting the damage boundary, damage coverage area, and damage expansion direction from the previous inspection time point and the current inspection time point, based on the time alignment results, to form the original damage change sequence corresponding to the previous inspection time point and the original damage change sequence corresponding to the current inspection time point; and using the Hampel filtering algorithm to compare the damage region expansion of the previous inspection time point and the current inspection time point. The original damage change sequence corresponding to the previous inspection point and the original damage change sequence corresponding to the current inspection point are processed by sliding window anomaly identification and anomaly replacement, respectively. Segmentation and aggregation processing is then performed on continuous anomaly locations to form the smooth damage change sequence corresponding to the previous inspection point, the smooth damage change sequence corresponding to the current inspection point, the anomaly segmentation result corresponding to the previous inspection point, and the anomaly segmentation result corresponding to the current inspection point. Local distance calculation is then performed on the smooth damage change sequences corresponding to the previous and current inspection points to generate a base distance matrix. Finally, the base distance matrix is ​​anomaly-adjusted based on the anomaly segmentation results corresponding to the previous and current inspection points. A constant position correction process is performed to generate a corrected distance matrix. This corrected distance matrix is ​​then combined with the abnormal segment results from the previous and current inspection points to perform a regularization window shrinkage process. The regularization window width is reduced for continuously changing segments, while a local elastic regularization window is retained for segments corresponding to abnormal segments, generating a regularized window result. Within this regularized window result, a step cost renormalization process is performed on the corrected distance matrix. The cumulative cost is increased for matching units within abnormal segments and those crossing abnormal segment boundaries, while the base cost is maintained for matching units within continuously changing segments, generating a step renormalization result. Finally, a forward regularization path search and a reverse regularization path search process are performed on the step renormalization result. The system generates forward regularization path results and reverse regularization path results. It then performs bidirectional path consistency verification on these results, identifying segments with consistent or adjacent path mappings as stable alignment segments and segments with deviated path mappings and corresponding abnormal segments as unstable alignment segments, generating path consistency results. Finally, it performs write-back and rollback processing on the unstable alignment segments in the path consistency results, combining the abnormal segment results from the previous inspection time point, the abnormal segment results from the current inspection time point, and the corresponding surface areas in the mixed tower digital twin data. Within the rolled-back local regularization window, it re-executes dynamic time regularization path search processing to generate write-back and re-regularization results.A unified integration process is performed on the stable alignment sections and the write-back re-normalization results. This is combined with the corresponding surface regions in the hybrid tower digital twin data to perform boundary extension correspondence processing, coverage area correspondence processing, expansion direction correspondence processing, and position consistency verification processing, generating extended comparison results.

[0027] The early warning output module receives damage time-series mapping data and extracts regional change content, tower part change content, and damage expansion content from the damage time-series mapping data. It performs regional change rate analysis processing on the regional change content to generate regional change rate results; it performs continuous change analysis processing on the tower part change content to generate continuous change results; it performs damage expansion direction analysis processing on the damage expansion content to generate damage expansion direction results; and it performs early warning result generation processing on the regional change rate results, continuous change results of tower parts, and damage expansion direction results, outputting damage evolution early warning data and hybrid tower digital twin visualization early warning results.

[0028] In this embodiment, the inspection image acquisition module receives images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, as well as the shooting pose information corresponding to each image, collected at different inspection times of the hybrid tower structure. First, the tower base area images are read one by one according to the inspection time points. Then, the concrete tower section images, steel-concrete connection section images, and upper tower section images are read according to the corresponding relationship of the same inspection time point. Simultaneously, the shooting pose information corresponding to each image is read, so that each image entering subsequent processing maintains a corresponding relationship with the corresponding shooting pose information.

[0029] When performing time-stamped association processing on images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section according to the inspection time point, the images of each tower base area are first assigned to the corresponding inspection time point. Then, the concrete tower section, steel-concrete connection section, and upper tower section images collected at the same inspection time point are respectively assigned to that inspection time point, and the inspection time point is written into the corresponding content of each image. Subsequently, the images at different inspection time points are arranged in the order of the inspection time points, so that each image always maintains a continuous correspondence with the corresponding inspection time point in the subsequent processing.

[0030] When performing spatial position association processing on the shooting pose information and corresponding images, the shooting pose information corresponding to each image is read one by one, and each shooting pose information is written into the associated position of the corresponding image. For images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, the corresponding shooting pose information and image content are synchronized and matched, so that each image carries the corresponding shooting pose information in the subsequent processing. For multiple images at the same inspection point, they are distinguished from each other according to their corresponding shooting pose information, so that different images still maintain a clear correspondence after spatial position association processing.

[0031] When performing part-to-part correspondence processing on image content and mixed tower structure parts, the image content of each image is read one by one and matched with the mixed tower structure parts; the image content corresponding to the tower base area is processed to the tower base area, the image content corresponding to the concrete tower section is processed to the concrete tower section, the image content corresponding to the steel-concrete connection section is processed to the steel-concrete connection section, and the image content corresponding to the upper tower section is processed to the upper tower section; when the image content crosses adjacent mixed tower structure parts, it is continuously processed according to the corresponding position of the image content in the mixed tower structure parts, so that the image content and the mixed tower structure parts maintain a continuous correspondence.

[0032] After completing the time-stamp association processing, spatial location association processing, and part correspondence processing, a unified integration process is performed on the images of the tower base area, concrete tower section, steel-concrete connection section, upper tower section, and the shooting pose information corresponding to each image. First, the images that have completed time-stamp association processing at the same inspection time point are integrated with the corresponding shooting pose information. Then, the correspondence that has completed spatial location association processing is written into the integration result. Subsequently, the image content that has completed part correspondence processing is written into the same integration result with the correspondence of the mixed tower structure parts. This ensures that different inspection time points, different image contents, corresponding shooting pose information, and corresponding mixed tower structure parts form a continuous connection in the unified integration process.

[0033] After unified integration, the images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, along with the corresponding shooting pose information, will be written according to the inspection time point to generate mixed tower inspection image data. The generated mixed tower inspection image data will retain the correspondence between the inspection time point, the shooting pose information, and the mixed tower structural parts, so that the subsequent unified view processing module can directly receive and perform image distortion correction, image scale unification, image view registration, and tower surface unfolding correlation processing.

[0034] In this embodiment, the unified view processing module receives mixed tower inspection image data and reads the tower base area image, concrete tower section image, steel-concrete connection section image, upper tower section image, and shooting pose information item by item from the mixed tower inspection image data. When reading the tower base area image, it maintains its correspondence with the corresponding inspection time point. When reading the concrete tower section image, steel-concrete connection section image, and upper tower section image, it maintains their respective correspondence with the corresponding shooting pose information, and ensures that the content of each image continues to maintain its correspondence with the mixed tower structure before entering subsequent processing.

[0035] When extracting images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, as well as shooting pose information, from mixed tower inspection image data, the tower base area image is first separated one by one according to the correspondence in the mixed tower inspection image data. Then, the concrete tower section image, steel-concrete connection section image, and upper tower section image are separated one by one according to the same correspondence method, and the shooting pose information corresponding to each image is extracted simultaneously. After the extraction is completed, each image is still associated with its own shooting pose information, and the original part correspondence of the tower base area image, concrete tower section image, steel-concrete connection section image, and upper tower section image remains unchanged.

[0036] When performing image distortion correction on images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, the image content of each image is read one by one, and distortion correction is performed according to the current image content of each image, so that the skew, stretching, and bending content in each image are adjusted accordingly. After completing the image distortion correction of one image, the same processing is performed on the next image, so that the images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section form the corrected image content before subsequent processing.

[0037] When performing image scale unification processing on images after image distortion correction, the image content of each image after image distortion correction is read one by one, and the scale of each image is adjusted according to a consistent image scale correspondence. After the scale adjustment is performed on the image of the tower base area, the corresponding scale adjustment is performed on the images of the concrete tower section, the steel-concrete connection section, and the upper tower section, so that the corresponding images of different mixed tower structure parts maintain a consistent image scale correspondence before entering the image viewpoint registration process, and the correspondence between each image and its respective shooting pose information is maintained.

[0038] When performing image viewpoint registration on images after image distortion correction and scale unification, the shooting pose information of each image is read one by one. Then, the current image content of each image is matched with the corresponding shooting pose information. After the position matching is completed, the images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section are adjusted according to the corresponding shooting pose information to form a corresponding viewpoint position relationship between different images. After the image viewpoint registration of each image is completed, the connection relationship between adjacent images is kept consistent with the correspondence between the corresponding shooting pose information.

[0039] When performing tower surface unfolding and correlation processing on the images after image viewpoint registration, the images of the tower base area, concrete tower segment, steel-concrete connection segment, and upper tower segment that have completed image viewpoint registration are read one by one, and unfolding correspondence is performed according to the corresponding concrete tower structure parts of each image; unfolding and correlation are performed on the image content in the tower base area image along the tower base area correspondence, unfolding and correlation are performed on the image content in the concrete tower segment image along the concrete tower segment correspondence, unfolding and correlation are performed on the image content in the steel-concrete connection segment image along the steel-concrete connection segment correspondence, and unfolding and correlation are performed on the image content in the upper tower segment image along the upper tower segment correspondence, so that the images corresponding to different concrete tower structure parts form a continuous tower surface correspondence after unfolding.

[0040] When performing association writing processing on the results after the tower surface unfolding association processing is completed, the tower base area image that has completed the tower surface unfolding association processing is first written into the unified result. Then, the concrete tower segment image, steel-concrete connection segment image, and upper tower segment image that have completed the tower surface unfolding association processing are sequentially written into the same result, and the shooting pose information corresponding to each image is synchronously written into the corresponding position. After the writing is completed, the image content, shooting pose information, image viewpoint registration relationship, and tower surface unfolding association relationship form a continuous connection relationship in the same result.

[0041] After completing the association writing process, unified view image data is generated. The generated unified view image data retains the correspondence between the tower base area image, concrete tower section image, steel-concrete connection section image, upper tower section image, and shooting pose information. It also retains the image content after image distortion correction, the scale correspondence after image scale unification, the view correspondence after image view registration, and the unfolding correspondence after tower surface unfolding association processing. This data is then used by the subsequent 3D twin modeling module to continue receiving and performing multi-view reconstruction processing, surface point cloud generation processing, tower surface mesh fitting processing, and texture mapping processing.

[0042] In this embodiment, the 3D twin modeling module receives unified viewpoint image data. After receiving the data, it reads the images of the tower base area, concrete tower section, steel-concrete connection section, upper tower section, and shooting pose information one by one from the unified viewpoint image data. It maintains the continuous correspondence between each image and the shooting pose information according to the correspondence already formed in the unified viewpoint image data, so that the image content entering subsequent processing continues to maintain the consistency with the concrete tower structure, the image viewpoint registration relationship, and the tower surface development association relationship.

[0043] When extracting image location content, image texture content, and surface unfolding location content from unified view image data, the spatial distribution content corresponding to each image is read item by item from each image to form image location content; then, the content related to the surface appearance of each image is read item by item from each image to form image texture content; subsequently, the corresponding positions after unfolding are read item by item from the correspondence after the tower surface unfolding association processing of each image to form surface unfolding location content; after the extraction is completed, the image location content, image texture content, and surface unfolding location content continue to maintain a homologous correspondence, and the content extracted from images corresponding to different mixed tower structure parts maintains a continuously stitchable correspondence.

[0044] When performing multi-view reconstruction on image location content, image texture content, and surface unfolded location content, the corresponding positions from different images are first compared item by item according to the adjacency relationship between the image location content. Then, the image texture content is synchronously written into the same reconstruction process with the corresponding image location content. The surface unfolded location content is used as the basis for continuous connection between each corresponding position, so that the different perspective content from the tower base area image, concrete tower section image, steel-concrete connection section image, and upper tower section image can be continuously stitched together under the same location frame. After completing item-by-item comparison, synchronous writing, and continuous stitching, the corresponding content of each perspective image forms a unified reconstruction correspondence, generating the reconstruction correspondence result.

[0045] When performing surface point cloud generation processing on the reconstructed corresponding results, the corresponding content of each position in the reconstructed corresponding results is read one by one, and the surface positions that different images point to are written into the point set according to the continuous distribution relationship of the corresponding content of each position in the reconstructed corresponding results. After completing the writing of one point, the writing of the corresponding positions adjacent to that point is continued, so that the corresponding content of each position gradually forms a continuously distributed point set according to the continuous relationship of the surface. After completing the writing of the corresponding content of all positions, a surface point cloud continuously corresponding to the outer surface of the hybrid tower structure is formed.

[0046] When performing tower surface mesh fitting processing on the reconstructed results, the set of adjacent points in the surface point cloud is first read. Then, the adjacent point sets are continuously connected according to the adjacency relationship between the points, so that the corresponding point sets of the tower base area, the corresponding point sets of the concrete tower section, the corresponding point sets of the steel-concrete connection section, and the corresponding point sets of the upper tower section form continuous surface connection relationships respectively. Subsequently, the connection positions between each continuous surface connection relationship are correspondingly sorted to keep the connection boundaries between different parts of the mixed tower structure continuous and consistent. After completing the continuous connection and corresponding sorting, the tower surface mesh fitting result corresponding to the outer surface of the mixed tower structure is formed.

[0047] When performing texture mapping processing on the results of the completed tower surface mesh fitting and the image texture content, the surface positions in the tower surface mesh fitting results are read one by one, and each surface position is matched with the previously extracted image texture content according to the correspondence. The surface positions corresponding to the tower base area are written with the image texture content corresponding to the tower base area, the surface positions corresponding to the concrete tower section are written with the image texture content corresponding to the concrete tower section, the surface positions corresponding to the steel-concrete connection section are written with the image texture content corresponding to the steel-concrete connection section, and the surface positions corresponding to the upper tower section are written with the image texture content corresponding to the upper tower section. After completing the matching, the surface mesh fitting results of the tower surface obtain a surface texture distribution consistent with each image, generating the three-dimensional surface model data of the hybrid tower.

[0048] When performing twin surface correspondence writing processing on the hybrid tower 3D surface model data, the surface positions, surface connection relationships, and surface texture correspondences in the hybrid tower 3D surface model data are read one by one. Then, according to the continuous surface relationships in the hybrid tower 3D surface model data, the surface positions are written sequentially into the same twin surface correspondence result. After the surface positions are written, the surface connection relationships and surface texture correspondences corresponding to each surface position are synchronously written into the corresponding positions, so that the same surface position always maintains a consistent connection with its surface connection relationships and surface texture correspondences. After all the writing is completed, a twin surface correspondence relationship consistent with the hybrid tower 3D surface model data is formed, generating hybrid tower digital twin data.

[0049] After generating the hybrid tower digital twin data, the hybrid tower digital twin data continues to maintain the surface correspondence with the hybrid tower 3D surface model data, and continues to maintain the source inheritance relationship with the image location content, image texture content and surface unfolding location content. It is also used continuously as the surface base data for subsequent temporal twin mapping module to perform spatial location inheritance processing and twin surface evolution writing processing.

[0050] In this embodiment, after receiving the unified view image data and the mixed tower 3D surface model data, the damage image analysis module first reads the image content of the crack area, spalling area, and corrosion area from the unified view image data image by image, and unfolds the crack area, spalling area, and corrosion area in each image sequentially according to the original image position relationship, so that the crack area, spalling area, and corrosion area maintain the correspondence with the original image position before entering the subsequent processing; then, it extracts the initial damage seed area from the unified view image data, and keeps each initial damage seed area synchronously corresponding to the original position in the corresponding image; after extracting the initial damage seed area, it continues to extract adjacent view images from the unified view image data, so that each initial damage seed area maintains the same corresponding base with the corresponding area in the adjacent view image; then, it reads the corresponding surface position corresponding to each initial damage seed area from the mixed tower 3D surface model data, and synchronously retrieves the corresponding surface position with the corresponding part position relationship of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, so that the initial damage seed area, adjacent view images, corresponding surface position, and part position relationship form the same processing chain before entering the seed area growth algorithm.

[0051] In the seed region growth algorithm process, firstly, each initial damage seed region is read boundary-by-boundary according to the boundary correspondence in adjacent view images, ensuring that each initial damage seed region has a corresponding boundary relationship with the adjacent view images. Then, based on the corresponding surface position in the 3D surface model data of the mixed tower, each initial damage seed region is read position-by-position, ensuring that each initial damage seed region has a positional correspondence with the corresponding surface position. Next, based on the corresponding positional relationships of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, each initial damage seed region is read part-by-part, ensuring that each initial damage seed region has a clear part-range attribution relationship. After completing the synchronous correspondence of boundary relationships, corresponding surface positions, and part-range positions, seed retention processing is performed on each initial damage seed region. Initial damage seed regions that maintain continuous boundary correspondence with adjacent view images, correspond to corresponding surface positions, and maintain consistent part-range positions are retained. Initial damage seed regions with interrupted boundary correspondence, deviated corresponding surface positions, or inconsistent part-range positions are stopped from retention, thus forming retained damage seed regions within their corresponding part ranges. Seed retention processing: ; ; : No. The overall retention judgment value of each initial damaged seed region : No. The correspondence between the initial damage seed region and the boundary in the adjacent viewpoint image is a continuous quantity. : No. The correspondence between an initial damage seed region and its corresponding surface location. : No. The consistency of the relationship between the initial damage seed region and the location of the site. : No. An initial damaged seed region, : No. The preserved damaged seed region Seed retention threshold Indicator function, which takes the value when the condition is true. When the condition is not met, the value is taken as... .

[0052] After forming the preserved damage seed region, when performing boundary expansion processing on the preserved damage seed region, firstly, pixels adjacent to the current boundary are read along the current boundary direction of the preserved damage seed region, and the continuity relationship of each adjacent pixel in grayscale variation is mapped one by one to the current damage seed region; then, the continuity relationship of each adjacent pixel in texture direction is mapped one by one to the current damage seed region, and the connection relationship of each adjacent pixel in boundary continuity is mapped one by one to the current damage seed region; finally, the correspondence of each adjacent pixel in location continuity is mapped one by one to the current damage seed region, so that each adjacent pixel has grayscale before entering the merging judgment. The relationships between pixels are considered: variation, texture direction, boundary continuity, and location continuity. After pixel-by-pixel mapping, adjacent pixels that maintain consistency with the current damage seed region in terms of grayscale variation, texture direction, boundary continuity, and location continuity are merged into the current damage seed region. Simultaneously, adjacent pixels that are discontinuous with the current damage seed region in terms of texture direction deviation are excluded from the current boundary, as are adjacent pixels that are discontinuous with the current damage seed region in terms of corresponding surface position offset. This ensures that the expanded boundary only extends outward along continuously connected positions. After merging and excluding pixels along each boundary direction, the expanded damage region boundary is formed. Boundary expansion processing: ; ; ; : No. The preserved damaged seed region has adjacent pixels at its current boundary. The continuous inclusion judgment value, Adjacent pixels The duration of grayscale changes in the currently damaged seed region. Adjacent pixels The amount of continuation of the current damaged seed region in the texture direction, Adjacent pixels The degree of continuity between the current damaged seed region and the boundary. Adjacent pixels The corresponding quantity in terms of locational continuity with the current damaged seed region. Adjacent pixels The exclusion criteria value Adjacent pixels The amount of deviation from the current damaged seed region in terms of texture orientation deviation. Adjacent pixels The offset relative to the current damaged seed region in terms of its position on the corresponding surface. : No. The current boundary of the preserved damaged seed region, : No. The boundary of the expanded damage area, : No. The set of neighboring pixels of the current boundary. Adjacent pixels are merged into the boundary value. Adjacent pixel exclusion threshold.

[0053] After forming the expanded damage region boundary, when performing feedback write-back processing on the expanded damage region boundary, firstly, the alignment result between the expanded damage region boundary and the corresponding boundary in the adjacent view image is read, ensuring that the current boundary maintains boundary alignment in the adjacent view image; then, the bonding result between the expanded damage region boundary and the corresponding surface position in the hybrid tower 3D surface model data is read, ensuring that the current boundary maintains surface bonding in the hybrid tower 3D surface model data; next, the continuity result between the expanded damage region boundary and the corresponding part position relationship is read, ensuring that the current boundary maintains part continuity within the corresponding part range; after completing the synchronous reading of alignment, bonding, and continuity results, continued expansion is performed on the continuous and stable boundary direction, allowing the current boundary to enter the next round of boundary expansion along the continuous and stable boundary direction; boundary regression is performed on the boundary direction with continuous offset and discontinuous part position relationship, allowing the current boundary to return to the boundary position before the continuous offset; after completing continued expansion and boundary regression, region update processing is performed on the updated boundary range, ensuring that the retained damage seed region corresponds again with the boundary range after boundary adjustment, forming damage segmentation result data. Feedback write-back processing and region update processing: ; ; ; ; : No. In the boundary of the expanded damage region, the first Write-back determination value for each boundary direction : No. The alignment result between each boundary direction and the corresponding boundary in the adjacent viewpoint image. : No. The amount of data showing the fit between each boundary direction and the corresponding surface location in the 3D surface model data of the hybrid tower. : No. A continuous result quantity relating each boundary direction to the positional relationship of the corresponding part. : No. Boundary correction amount in each boundary direction. : No. Correction step size in each boundary direction, The threshold for determining whether the boundary continues to expand or regresses. : No. The overall boundary correction set of the expanded damage region boundary : No. The updated boundary range, Damage segmentation results data, The area enclosed by the updated boundary. Set union operation.

[0054] After receiving the damage segmentation result data and the hybrid tower 3D surface model data, the damage 3D projection module first reads damage boundary segments from different viewpoints in the damage segmentation result data, maintaining the correspondence between each damage boundary segment and the original viewpoint image. Then, it reads the surface position relationships corresponding to the tower base region, concrete tower segment, steel-concrete connection segment, and upper tower segment from the hybrid tower 3D surface model data, ensuring that each damage boundary segment has a corresponding surface position foundation before entering the projection process. After completing the synchronous reading of damage boundary segments and surface position relationships, a seed region growing algorithm is used to perform damage region pose projection processing on the damage segmentation result data and the hybrid tower 3D surface model data, writing each damage boundary segment into its corresponding position in the hybrid tower 3D surface model data according to its original pose relationship. After completing the damage region pose projection processing, based on the surface position relationships corresponding to the tower base region, concrete tower segment, steel-concrete connection segment, and upper tower segment, trimming processing is performed on damage boundary segments projected to the same area and with adjacent projection positions. Damage region pose projection processing: ; ; ; : No. The first damage boundary segment A boundary point, : No. The pose rotation matrix corresponding to each damaged boundary segment. : No. The pose translation vector corresponding to each damaged boundary segment. : No. The first damage boundary segment The pose projection points of each boundary point : No. The first damage boundary segment The surface landing points of each boundary point in the 3D surface model data of the hybrid tower. : No. The set of surface locations within the area corresponding to each damage boundary segment. : No. A projected damage boundary segment, : No. The number of boundary points in each damaged boundary segment : Norm 2 distance.

[0055] In the trimming correspondence processing, firstly, segment-by-segment positional correspondence is performed on damage boundary segments projected to the same location range, forming the same comparison range for damage boundary segments within the same location range; then, segment-by-boundary orientation correspondence is performed on damage boundary segments within the same comparison range and with adjacent projection positions, forming a segment-by-segment comparison relationship in terms of boundary orientation; next, segment-by-segment consistency correspondence is performed on the positional relationship between each damage boundary segment and its corresponding location, forming a segment-by-segment comparison relationship in terms of location position; after completing the projection position correspondence, boundary orientation correspondence, and location positional relationship correspondence, correspondences inconsistent with the current projection position, boundary orientation, and location position are eliminated, leaving only damage boundary segments with continuous projection position, continuous boundary orientation, and continuous location position, forming the damage boundary segments after trimming correspondence processing. Trimming correspondence processing: ; ; : No. The projected damage boundary segment and the first Trim consistency criteria among projected damage boundary segments. The consistency of the two damage boundary segments at their projected locations. The consistency of the two damaged boundary segments along the boundary orientation. The degree of consistency in the locational relationship between two damage boundary segments. : No. A projected damage boundary segment, : No. Each damaged boundary segment after trimming has been processed. : Trim the corresponding retention threshold, Existence and the first One damage boundary segment constitutes another damage boundary segment in the comparison relationship.

[0056] After forming the trimmed and corresponding damage boundary segments, when performing point-to-surface convergence processing on the trimmed and corresponding damage boundary segments and the local surface segments in the hybrid tower 3D surface model data, the bonding results of each damage boundary segment to the local surface normal direction are read segment by segment first, so that each damage boundary segment has a bonding relationship with the local surface normal direction; then, the positional continuity relationship between the local surface segments is read, so that each local surface segment forms a positional continuity relationship within the same part; then, the surface connection relationship within the corresponding part is read, so that each local surface segment forms a surface continuity relationship within the corresponding part; after completing the synchronous reading of bonding results, positional continuity relationship and surface connection relationship, the projection position of each damage boundary segment in the hybrid tower 3D surface model data is adjusted according to the bonding results of each damage boundary segment to the local surface normal direction, the relative position between each damage boundary segment is corrected according to the positional continuity relationship between the local surface segments, and the connection position of each damage boundary segment on the surface is corrected according to the surface connection relationship within the corresponding part, so that each damage boundary segment forms a continuous bonding relationship on the local surface segment, forming the updated damage boundary segment. Point-to-surface convergence processing: ; ; ; : No. The result of fitting the damaged boundary segment, after trimming and corresponding processing, to the local surface normal direction. : No. The unit vector of the normal direction of a local surface segment. : No. The current projection position of the corresponding damaged boundary segment after trimming. : with the The center position of the local surface segment corresponding to the damaged boundary segment after trimming. : No. The location of the updated damage boundary segment The positional correction amount is obtained based on the positional continuity relationship between local surface segments. The surface correction amount is obtained based on the surface connection relationship within the corresponding part range. Normal fit correction factor : Position continuity correction factor : Correction coefficient for surface connection relationship : No. An updated damage boundary segment, : Transpose operation.

[0057] After forming the updated damage boundary segments, when performing spatial merging processing on each updated damage boundary segment, firstly, each damage boundary segment is matched segment by segment based on its spatial location, so that damage boundary segments with continuous spatial locations enter the same merging range; then, each segment is matched segment by segment based on its boundary orientation, so that damage boundary segments with continuous boundary orientations form a boundary connection relationship; next, each segment is matched segment by segment based on its surface bonding result, so that damage boundary segments with continuous surface bonding results form a surface continuity relationship; finally, the damage boundary segments are matched segment by segment based on their location relationship. Damage boundary segments within the same merging range undergo segment-by-segment location mapping to ensure continuous location relationships among them. After completing the item-by-item mapping of spatial location, boundary orientation, surface fitting results, and location relationships, the continuous and consistent damage boundary segments are merged into the same damage region. After merging the same damage regions, a duplicate mapping reduction process is performed on the merged damage region. This process removes duplicate portions from the mixed-tower 3D surface model data, retaining only the single projection region corresponding to the damage region within the mixed-tower 3D surface model data, thus forming the 3D projection data of the damage. Spatial merging of homogeneous damage: ; ; ; : No. The updated damage boundary fragment and the first The merge consistency judgment value among the updated damage boundary segments. : The quantity of spatial continuity between two damage boundary segments : The continuity and consistency of the two damaged boundary segments along the boundary orientation. The continuity and consistency of the two damaged boundary segments in the surface bonding result. The continuity and consistency of the two damage boundary segments in terms of their location relationship. : No. A set of damage boundary segment numbers corresponding to the same damage region. : No. The same damaged area, : No. An updated damage boundary segment, Merge decision threshold. Duplicate mapping reduction processing: ; ; ; : No. The retention criteria value for the same damaged area : No. The consistency of the same damaged area at the projection location : No. Consistent quantity of the same damaged areas along their boundary orientation : No. The consistency in the location relationship of the same damaged areas : with the A set of identical damage regions that overlap. The single projected region that remains after duplicate mapping reduction. Damage 3D projection data, : Empty set.

[0058] In this embodiment, after receiving the damage 3D projection data and the mixed tower digital twin data, the time-series twin mapping module first reads the damage 3D projection data one by one according to the inspection time point, so that the damage 3D projection data corresponding to different inspection time points are arranged in chronological order; then, the damage 3D projection data corresponding to adjacent inspection time points are matched sequentially according to the inspection time point order, so that the damage 3D projection data corresponding to the previous inspection time point and the damage 3D projection data corresponding to the current inspection time point form a group correspondence relationship; subsequently, time alignment processing is performed on the damage 3D projection data corresponding to different inspection time points that form a group correspondence relationship, so that a one-to-one time connection relationship is formed between the previous inspection time point and the current inspection time point, generating a time alignment result; after forming the time alignment result, the time alignment result and the mixed tower digital twin data are matched according to the spatial position relationship, so that the damage 3D projection data corresponding to the previous inspection time point and the damage 3D projection data corresponding to the current inspection time point form a successor relationship on the corresponding surface area in the mixed tower digital twin data, generating a spatial inheritance result. The time alignment result formula is: ; ; Compared with the previous inspection time point The location of the current inspection time point sequence that is closest in time to the three-dimensional projection data of the damage. : No. The time alignment results : Previous inspection time point Three-dimensional projection data of the damage, Compared with the previous inspection time point The damage 3D projection data is completed after time alignment of the current inspection time point damage 3D projection data. : Previous inspection time point Inspection time for each damage 3D projection data Current inspection time point Inspection time for each damage 3D projection data. Spatial inheritance result formula: ; ; : No. The corresponding surface region location of each time alignment result in the hybrid tower digital twin data. : No. The result of spatial inheritance, : No. The time alignment results : Previous inspection time point The corresponding surface location of each damage 3D projection data in the hybrid tower digital twin data. The corresponding surface location in the hybrid tower digital twin data of the damage 3D projection data after time alignment at the current inspection point. : The set of corresponding surface regions in the hybrid digital twin data.

[0059] After generating the spatial inheritance results, a dynamic time warping algorithm combined with a Hampel filtering algorithm is used to perform damage region expansion comparison processing on the spatial inheritance results, time alignment results, and mixed tower digital twin data. During the damage region expansion comparison processing, the 3D projection data of the damage corresponding to the previous inspection time point and the 3D projection data of the damage corresponding to the current inspection time point in the spatial inheritance results are first processed according to their spatial positional relationship to form a continuous region correspondence, ensuring that the 3D projection data of the damage corresponding to the previous inspection time point and the 3D projection data of the damage corresponding to the current inspection time point, which are under the same spatial positional relationship, form a continuous correspondence. Then, combined with the time alignment results, the damage boundary, damage coverage area, and damage expansion direction at the previous inspection time point are extracted position by position from the 3D projection data of the damage corresponding to the previous inspection time point, and the damage boundary, damage coverage area, and damage expansion direction at the current inspection time point are also extracted. The damage boundary, damage coverage area, and damage propagation direction are extracted position-by-position from the corresponding 3D damage projection data at the current inspection time point, ensuring that the extraction of these parameters is synchronized between two consecutive inspection time points. After synchronized extraction, the damage boundary, damage coverage area, and damage propagation direction from the previous inspection time point are continuously written in chronological order to form the original damage change sequence corresponding to the previous inspection time point. Similarly, the damage boundary, damage coverage area, and damage propagation direction from the current inspection time point are continuously written in chronological order to form the original damage change sequence corresponding to the current inspection time point. The formula for the original damage change sequence corresponding to the previous inspection time point is: Formula for the original damage change sequence corresponding to the current inspection point: ; : Previous inspection time point One original damage change sequence unit, Current inspection time point One original damage change sequence unit, : Previous inspection time point A damage boundary, : Previous inspection time point The amount of damage coverage area, : Previous inspection time point One direction of damage propagation, The current inspection time point after time alignment A damage boundary, The current inspection time point after time alignment The amount of damage coverage area, The current inspection time point after time alignment One direction of damage propagation.

[0060] After generating the original damage change sequence corresponding to the previous inspection time point and the original damage change sequence corresponding to the current inspection time point, the Hampel filtering algorithm is used to perform sliding window anomaly identification and anomaly replacement processing on the original damage change sequence corresponding to the previous inspection time point. During sliding window anomaly identification, the sequence position content of each position in the original damage change sequence corresponding to the previous inspection time point is read sequentially according to the sequence position, ensuring that the current reading position maintains a continuous window relationship with its preceding and following positions. Then, the current reading position is compared with other positions within the window to establish anomaly comparison relationships between the current reading position and continuous positions within the window. Subsequently, anomaly identification processing is performed based on these anomaly comparison relationships, creating anomaly position relationships for positions deviating from the continuous change relationship. After the anomaly position relationships are formed, anomaly replacement processing is performed on the anomaly positions, replacing them with replacement results that maintain consistency with the continuous position relationships within the window. When multiple consecutive anomaly positions appear, segmentation and aggregation processing is performed on the consecutive anomaly positions, forming continuous anomaly segments between adjacent anomaly positions, generating the smoothed damage change sequence corresponding to the previous inspection time point and the anomaly segmentation results corresponding to the previous inspection time point. When using the Hampel filtering algorithm to perform sliding window anomaly identification and replacement processing on the original damage change sequence corresponding to the current inspection time point, the same processing order as the original damage change sequence corresponding to the previous inspection time point is followed. This involves position-by-position reading, difference comparison, anomaly identification, anomaly replacement, and segmentation and aggregation of continuous anomaly positions within each sequence position in the original damage change sequence corresponding to the current inspection time point. This generates the smoothed damage change sequence corresponding to the current inspection time point and the anomaly segmentation results corresponding to the current inspection time point. Sliding window value: Sliding window deviation: Smooth damage change sequence: Abnormal location marker: Comprehensive anomaly location marker: Abnormal segmentation results: ; Current inspection time point The first original damage change sequence unit One portion, Current inspection time point The sequence position of the th sequence position Values ​​in a sliding window of each component Current inspection time point The sequence position of the th sequence position The sliding window deviation of each component Current inspection time point The sequence position of the th sequence position Smoothing results for each component, Current inspection time point The sequence position of the th sequence position Anomaly location markers for each component, Current inspection time point Comprehensive anomaly location markers for each sequence position. The current inspection point corresponds to the abnormal segment results. : The number of the variable component, where Represents the damage boundary component. Indicates the component of the damaged area. Indicates the component of the damage propagation direction. Sliding window half-width Anomaly detection amplification factor : The start and end numbers of the abnormal segments formed by continuous abnormal locations, Replace with This will yield the smoothed damage change sequence corresponding to the previous inspection time point and the abnormal segmentation results corresponding to the previous inspection time point. By keeping the current inspection time point as the reference, we can obtain the smoothed damage change sequence and the abnormal segmentation results corresponding to the current inspection time point.

[0061] After generating the smooth damage change sequence corresponding to the previous inspection time point, the smooth damage change sequence corresponding to the current inspection time point, the abnormal segmentation result corresponding to the previous inspection time point, and the abnormal segmentation result corresponding to the current inspection time point, the smooth damage change sequence corresponding to the previous inspection time point and the smooth damage change sequence corresponding to the current inspection time point are processed by position-by-position local distance calculation. When performing position-by-position local distance calculation, firstly, the contents of each position in the smooth damage change sequence corresponding to the previous inspection time point are read in sequence position order, and simultaneously, the contents of each position in the smooth damage change sequence corresponding to the current inspection time point are read. Then, position-by-position correspondence calculation is performed between the current position contents in the smooth damage change sequence corresponding to the previous inspection time point and the current position contents in the smooth damage change sequence corresponding to the current inspection time point, establishing local distance relationships between corresponding positions. Subsequently, all local distance relationships are written into the base distance matrix in the position order of the smooth damage change sequence corresponding to the previous inspection time point and the current inspection time point, generating the base distance matrix. After forming the base distance matrix, anomaly position correction processing is performed on the base distance matrix based on the anomaly segmentation results corresponding to the previous and current inspection times, establishing anomaly correction relationships between the matrix positions in the anomaly segmentation results corresponding to the previous and current inspection times, and these anomaly correction relationships are written into the corresponding positions in the base distance matrix, generating the corrected distance matrix. Position-by-position local distance calculation: ; The first in the basic distance matrix Line number Local distance values ​​of the column, : Previous inspection time point The damage boundary and the current inspection time point The boundary distance between each damage boundary : Previous inspection time point The amount of damage coverage area, Current inspection time point The amount of damage coverage area, : Previous inspection time point One direction of damage propagation, Current inspection time point One direction of damage propagation, Damage boundary distance weight, Weight of the difference in the amount of damage covered. Damage propagation direction difference weight. Abnormal location correction marker: Corrected distance matrix: ; The first in the basic distance matrix Line number Anomaly correction markers at the corresponding column positions, : The corrected distance matrix Line number Corrected distance value of the column, The first in the basic distance matrix Line number Local distance values ​​of the column, The abnormal segment results corresponding to the previous inspection point. The current inspection point corresponds to the abnormal segment results. : Incremental coefficient for abnormal location correction.

[0062] After forming the corrected distance matrix, a regularized window shrinking process is performed on the corrected distance matrix, combining the abnormal segment results corresponding to the previous inspection time point and the abnormal segment results corresponding to the current inspection time point. During the regularized window shrinking process, the corresponding positions of continuously changing segments and abnormal segments in the corrected distance matrix are first read; then, the width of the regularized window for the corresponding positions of continuously changing segments is reduced, causing the path search range for these positions to converge towards the continuous corresponding direction; subsequently, a local flexible regularized window is retained for the corresponding positions of abnormal segments, ensuring that the path search range for these positions remains locally adjustable; after completing the different window arrangements for the corresponding positions of continuously changing segments and abnormal segments, the regularized window result is generated. Regularized window result: ; The first result in the regularized window Line number The column's position is a feasible marker. The regular window width used for continuously changing sections. The width of the local flexible regularization window used for the segment corresponding to the abnormal segmentation. The abnormal segment results corresponding to the previous inspection point. The current inspection point corresponds to the abnormal segment results. : No. Each sequence position does not fall within the abnormal segment interval of the previous inspection point. : No. Each sequence position does not fall within the abnormal segment interval of the current inspection time point. : No. The sequence position falls within the abnormal segment interval of the previous inspection point. : No. The sequence position falls within the abnormal segment interval of the current inspection time point.

[0063] After generating the regularized window result, step cost renormalization is performed on the corrected distance matrix within the regularized window result. During step cost renormalization, the corrected distance matrix positions corresponding to the step positions of each path in the regularized window result are first read; then, the cumulative cost of matching units within outlier segments is increased, giving them a higher cumulative cost when entering path search; subsequently, the cumulative cost of matching units crossing outlier segment boundaries is increased, giving them a higher cumulative cost when entering path search; then, the base cost of matching units within continuously changing segments is maintained, ensuring that the matching units within continuously changing segments retain the original cost relationships in the corrected distance matrix; after completing the corresponding adjustments for matching units within outlier segments, matching units crossing outlier segment boundaries, and matching units within continuously changing segments, the step renormalization result is generated. Cross-outlier segment boundary marker: Step-by-step reorganization results: ; : No. Line number A marker indicating whether a column position crosses an abnormal segment boundary. The first step in the reorganization results Line number The progressive value of the column : The corrected distance matrix Line number Corrected distance value of the column, Abnormal location correction marker, : The cumulative cost increase coefficient of matching units within abnormal segments. : The cumulative cost increase coefficient for matching units that cross abnormal segment boundaries.

[0064] After generating the step-down reshaping result, forward regularization path search and reverse regularization path search are performed on the result. During forward regularization path search, the cumulative cost in the step-down reshaping result is gradually accumulated along the allowed path direction in the regularization window, following the order from the smoothed damage change sequence corresponding to the previous inspection point to the smoothed damage change sequence corresponding to the current inspection point. This creates a forward cumulative relationship between the candidate paths from the start to the end position, and the path with the lowest consecutive cumulative cost is retained from these relationships to generate the forward regularization path result. During reverse regularization path search, the cumulative cost in the step-down reshaping result is gradually accumulated along the allowed path direction in the regularization window, following the order from the smoothed damage change sequence corresponding to the current inspection point to the smoothed damage change sequence corresponding to the previous inspection point. This creates a reverse cumulative relationship between the candidate paths from the start to the end position, and the path with the lowest consecutive cumulative cost is retained from these relationships to generate the reverse regularization path result. Forward cumulative cost matrix: Forward regularization path results: ; ; The first digit in the forward cumulative cost matrix Line number The forward cumulative cost of the column, The first step in the reorganization results Line number The progressive value of the column The first result in the regularized window Line number The column's position is a feasible marker. : The forward regularization path and the previous inspection point The current inspection point path position corresponding to each sequence position. : Forward regularized path result : Length of the smoothed damage change sequence corresponding to the previous inspection point. Inverse cumulative cost matrix: ; Reverse normalization path results: ; ; The th in the reverse cumulative cost matrix Line number The reverse cumulative cost of the column, The first step in the reorganization results Line number The progressive value of the column The first result in the regularized window Line number The column's position is a feasible marker. : In the reverse normalization path, the point at the previous inspection time is The current inspection point path position corresponding to each sequence position. : Reverse normalization path results : The length of the smoothed damage change sequence corresponding to the previous inspection point.

[0065] After generating the forward and reverse regularization path results, a bidirectional path consistency check is performed on both. During this check, the path mapping relationships in the forward regularization path results are read sequentially by path position, and simultaneously, the path mapping relationships in the reverse regularization path results are read. Then, the path mapping relationships in the forward and reverse regularization path results are compared segment by segment to ensure consistency within the same segment. Segments with consistent or adjacent path mapping relationships are then identified as stable alignment segments, maintaining continuous path mapping relationships. Segments with deviated path mapping relationships and corresponding abnormal segments are identified as unstable alignment segments, maintaining abnormal path mapping relationships. After determining the stable and unstable alignment segments, a path consistency result is generated. Path position consistency determination: Path consistency results: Stable alignment segments and unstable alignment segments: ; ; : No. Consistency markers for each path location, Path consistency results The first in the forward regularized path Mapping position of the current inspection time point for each path location : The first in the reverse regularization path Mapping position of the current inspection time point for each path location : The set of positions corresponding to the stable alignment segment The set of positions corresponding to unstable alignment segments. The abnormal segment results corresponding to the previous inspection point. : The abnormal segment result corresponding to the current inspection time point.

[0066] After the path consistency results are generated, the unstable alignment sections in the path consistency results are combined with the abnormal segment results corresponding to the previous inspection time point, the abnormal segment results corresponding to the current inspection time point, and the corresponding surface areas in the mixed tower digital twin data to perform write-back and rollback processing. During the write-back and rollback process, the corresponding positions of the unstable alignment segments in the abnormal segment results at the previous inspection time point are first read, and then the corresponding positions of the unstable alignment segments in the abnormal segment results at the current inspection time point are read, establishing a synchronous correspondence between the unstable alignment segments and the abnormal segment results at the previous and current inspection times. Next, the corresponding surface areas of the unstable alignment segments in the hybrid tower digital twin data are read, establishing a positional correspondence between the unstable alignment segments and their corresponding surface areas. Subsequently, the write-back and rollback process is performed on the unstable alignment segments, causing them to revert from their current path mapping position to the previous position where a continuous correspondence is maintained. After the write-back and rollback process is completed, the dynamic time-normalized path search process is re-executed within the rolled-back local normalization window, causing the rolled-back local segments to re-form a local path mapping relationship, generating a write-back re-normalization result. The rolled-back local normalization window is shown below. Write back and reorganize the results: ;

[0067] : No. The set of possible path positions for an unstable alignment position within a locally normalized window after rollback. : No. The path mapping position after write-back, rollback and local realignment. Write back and reorganize the results. : The set of positions corresponding to the stable alignment segment The set of positions corresponding to unstable alignment segments. : The corrected distance matrix Line number Corrected distance value of the column, The first in the forward regularized path Mapping location of each path location.

[0068] After generating the write-back and re-normalization results, a unified integration process is performed on the stable alignment segments and the write-back and re-normalization results. During this process, the stable alignment segments are first written into the unified path chain in their original path order; then, the write-back and re-normalization results are written into the corresponding segments in the unified path chain in their corresponding position order, ensuring a continuous connection between the stable alignment segments and the write-back and re-normalization results. After completing the unified integration process, boundary extension correspondence processing is performed on the corresponding surface areas in the hybrid tower digital twin data, ensuring a continuous boundary correspondence between the damage boundaries in the previous inspection point and the damage boundaries in the current inspection point on the corresponding surface areas. Finally, coverage correspondence processing is performed, ensuring a continuous boundary correspondence between the damage coverage areas in the previous inspection point and the damage coverage areas in the current inspection point. The damaged coverage area forms a continuous coverage correspondence on the corresponding surface area; then, expansion direction correspondence processing is performed to ensure that the damage expansion direction in the previous inspection point and the damage expansion direction in the current inspection point form a continuous expansion direction correspondence on the corresponding surface area; next, position consistency verification processing is performed to ensure that the results of boundary extension correspondence processing, coverage area correspondence processing, and expansion direction correspondence processing maintain positional consistency with the corresponding surface area in the hybrid tower digital twin data; after completing boundary extension correspondence processing, coverage area correspondence processing, expansion direction correspondence processing, and position consistency verification processing, expansion comparison results are generated. Unified path mapping: ; Boundary extension corresponding quantity: Coverage area corresponding quantity: ; Quantities corresponding to the expansion direction: Position consistency quantity: Extended alignment results: ; ; The unified and integrated path mapping location. : No. The corresponding amount of boundary extension at each sequence position. : No. The coverage area corresponding to each sequence position, : No. The quantity corresponding to the expansion direction of each sequence position. : No. The positional consistency of each sequence position : No. The expanded alignment result value of each sequence position, Expand the set of alignment results. : The distance between the damage boundary of the previous inspection point and the current inspection point under the unified path mapping. : Normalized upper bound of damage boundary distance : Normalized upper bound of the difference in the amount of damage covered area : Normalized upper bound of the difference in damage propagation direction : The normalized upper bound of the corresponding surface position difference. : Boundary extension corresponds to the weight of the quantity. Coverage range corresponds to the weight of the quantity. : The corresponding weight of the expansion direction Position consistency weight.

[0069] After generating the extended alignment results, a twin surface evolution writing process is performed on them. During this process, the boundary extension correspondence, coverage correspondence, and expansion direction correspondence in the extended alignment results are first read. Then, the boundary extension correspondence is written into the corresponding surface region in the hybrid tower digital twin data, ensuring a continuous evolution of the boundary relationships within the corresponding surface region as the inspection time changes. Next, the coverage correspondence is written into the corresponding surface region in the hybrid tower digital twin data, ensuring a continuous evolution of the coverage relationship as the inspection time changes. Finally, the expansion direction correspondence is written into the corresponding surface region in the hybrid tower digital twin data, ensuring a continuous evolution of the expansion direction relationship as the inspection time changes. After completing the twin surface evolution writing process for the boundary extension correspondence, coverage correspondence, and expansion direction correspondence, damage time-series mapping data is generated. Twin surface evolution writing results: Damage time-series mapping data: ; : No. The twin surface evolution writing results on each corresponding surface region Damage time-series mapping data The location of the corresponding surface area under the unified path mapping at the current inspection time. : No. The corresponding amount of boundary extension at each sequence position. : No. The coverage area corresponding to each sequence position, : No. The quantity corresponding to the expansion direction of each sequence position. : No. The expanded alignment result value of each sequence position.

[0070] In this embodiment, after receiving the damage time-series mapping data, the early warning output module first reads the data content of each time point in the damage time-series mapping data according to the inspection time point order, so that the data content corresponding to the inspection time points before and after maintains the time sequence relationship; then, it extracts the regional change content, tower part change content, and damage expansion content from the damage time-series mapping data in sequence, so that the regional change content, tower part change content, and damage expansion content maintain the corresponding relationship under the same inspection time point, and uses them as input content for subsequent analysis and processing.

[0071] When performing regional change rate analysis on regional change content, the system first reads the current time point regional change record and the previous time point regional change record in the regional change content hourly according to the inspection time point sequence, so that the regional change records in adjacent inspection time points form a corresponding relationship; then, the current time point regional change record and the previous time point regional change record are compared according to the same spatial position relationship, so that the regional change situation in the previous and previous inspection time points under the same spatial position forms a continuous comparison relationship; then, based on the regional change situation in the previous and previous inspection time points, the continuous change amplitude and the relationship between the two changes are determined, and the regional change amplitude corresponding to each inspection time point is continuously organized in chronological order to generate the regional change rate result.

[0072] When performing continuous change analysis on tower body parts, the process first reads the tower body part change records at the current and previous times in the order of inspection times, ensuring a correspondence between tower body part change records at adjacent inspection times. Then, the current and previous tower body part change records are compared according to the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, ensuring a continuous comparison of changes within the same area at different inspection times. Finally, based on the changes within the same area at different inspection times, the continuity and succession relationships of tower body part changes are determined, generating continuous change results for tower body parts.

[0073] When performing damage expansion direction analysis on the damage expansion content, the damage expansion record at the current time point and the damage expansion record at the previous time point are read one by one in the order of the inspection time points to form a front-to-back correspondence between the damage expansion records at adjacent inspection time points; then, the damage expansion record at the current time point and the damage expansion record at the previous time point are compared according to the same spatial position relationship to form a continuous correspondence between the expansion situation at the previous and previous inspection time points under the same spatial position; finally, the continuous change relationship and front-to-back direction relationship of the damage expansion direction are determined based on the change of expansion position at the previous and previous inspection time points to generate the damage expansion direction result.

[0074] When generating early warning results for the regional change rate results, the continuous change results of tower body parts, and the damage propagation direction results, the following steps are taken: First, the regional change rate results, the continuous change results of tower body parts, and the damage propagation direction results are read synchronously according to the inspection time point sequence, ensuring that the three types of results at the same inspection time point maintain a synchronous correspondence. Then, the regional change rate results, the continuous change results of tower body parts, and the damage propagation direction results are organized according to their spatial location, forming a unified correspondence between the regional change situation, the tower body part change situation, and the damage propagation direction situation at the same spatial location. Subsequently, early warning results corresponding to each inspection time point are generated based on the unified correspondence, and damage evolution early warning data and hybrid tower digital twin visualization early warning results are formed. After the generation is completed, the damage evolution early warning data and hybrid tower digital twin visualization early warning results are output.

[0075] Example 1: In this example, the inspection image acquisition module receives images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section at the same inspection time point, as well as the shooting pose information corresponding to each image. It maintains the time sequence of the four types of images according to the inspection time point, maintains the spatial position relationship of each image according to the shooting pose information, and maintains the part correspondence of the image content according to the mixed tower structure. Then, it continuously writes the corresponding image content and shooting pose information completed at the same inspection time point into the same data chain to generate mixed tower inspection image data. After receiving the mixed tower inspection image data, the unified view processing module sequentially performs image distortion correction, image scale unification, image view registration, and tower surface unfolding association processing on the tower base area image, concrete tower section image, steel-concrete connection section image, and upper tower section image within the same inspection time point. This ensures that the images of different parts are continuously unfolded along the surface of the mixed tower structure under a unified scale and unified view relationship. The unfolded image content is synchronously written with the shooting pose information to generate unified view image data. After receiving image data from a unified perspective, the 3D twin modeling module continuously extracts image location content, image texture content, and surface unfolding location content from the image content corresponding to the same inspection time point. Then, based on the correspondence between these contents, it completes multi-view reconstruction processing, surface point cloud generation processing, tower surface mesh fitting processing, and texture mapping processing. This allows the surface textures of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section to be continuously spliced ​​under the same surface framework, generating mixed tower 3D surface model data. Based on the mixed tower 3D surface model data, mixed tower digital twin data is generated. After receiving image data from a unified perspective, the damage image analysis module performs crack region segmentation, spalling region segmentation, and corrosion region segmentation, as well as damage edge contour refinement, around the crack, spalling, and corrosion regions. This ensures that the damage boundaries in different parts of the mixed tower structure maintain a continuous recognition relationship at their corresponding image positions, forming damage segmentation result data. The damage 3D projection module continues to receive the damage segmentation result data and the mixed tower 3D surface model data. It performs damage region pose projection, surface landing point correspondence processing, homogeneous damage space merging processing, and repetitive mapping reduction processing on the damage boundary fragments from images from different perspectives. This merges the boundary fragments corresponding to the same damage region under different perspectives into a single projection region in the mixed tower 3D surface model data, generating damage 3D projection data. After receiving the damage 3D projection data and mixed tower digital twin data corresponding to different inspection time points, the temporal twin mapping module performs time alignment processing and spatial position inheritance processing on the damage 3D projection data between the previous and subsequent inspection time points, so that the damage boundary, damage coverage area and damage expansion direction at the same spatial location form a succession relationship between different inspection time points. Then, based on the succession relationship, damage area expansion comparison processing and twin surface evolution writing processing are performed to generate damage temporal mapping data. After receiving the damage time-series mapping data, the early warning output module extracts the changes in the region, the changes in the tower body, and the damage expansion according to the inspection time point. It then continuously performs regional change rate analysis, continuous change analysis of the tower body, damage expansion direction analysis, and early warning result generation around the same spatial location. This unifies the changes in the tower base area, concrete tower section, steel-concrete connection section, and upper tower section at different inspection time points into the same early warning output chain, outputting damage evolution early warning data and visualized early warning results of the hybrid tower digital twin.

[0076] Example 2: In this example, after the damage image analysis module receives the unified view image data and the three-dimensional surface model data of the mixed tower, it first continuously extracts the initial damage seed region and adjacent view images around the original image positions of the crack area, the peeling area and the corrosion area in the unified view image data. Then, it keeps the initial damage seed region synchronously corresponding to the corresponding surface position in the three-dimensional surface model data of the mixed tower, as well as the corresponding part position relationship of the tower base area, the concrete tower section, the steel-concrete connection section and the upper tower section, so that the initial damage seed region has the image position basis, the surface position basis and the part position basis before entering the subsequent processing. After the initial damage seed region is synchronized, the damage image parsing module continuously performs seed preservation processing, boundary expansion processing and damage edge contour refinement processing around the same initial damage seed region. This ensures that the boundary range is continuously preserved and extrapolated, maintaining continuity with the boundary correspondence in adjacent viewpoint images, adhering to the corresponding surface position in the mixed tower 3D surface model data, and maintaining consistency with the positional relationship of the corresponding parts, thus forming damage segmentation result data. After receiving the damage segmentation result data and the mixed tower 3D surface model data, the damage 3D projection module performs damage region pose projection processing and surface landing point correspondence processing around the damage boundary segments from images from different perspectives, so that each damage boundary segment enters the corresponding position in the mixed tower 3D surface model data along the original pose relationship; then, it continues to perform homogeneous damage spatial merging processing and repeated mapping reduction processing, so that damage boundary segments with continuous spatial position, continuous boundary direction, continuous surface fitting result and continuous location relationship are merged into the same damage region, generating damage 3D projection data. After receiving the damage 3D projection data and the mixed tower digital twin data, the temporal twin mapping module first performs time alignment processing on the damage 3D projection data corresponding to the previous inspection time point and the damage 3D projection data corresponding to the current inspection time point according to the inspection time point. Then, it performs corresponding organization on the time alignment result and the mixed tower digital twin data according to the spatial position relationship, so that the damage areas before and after the same spatial position form a spatial position inheritance relationship. On this basis, it extracts the damage boundary, damage coverage area and damage expansion direction in the previous inspection time point and the current inspection time point to form a continuous comparison basis for the inspection time points before and after. After the continuous comparison base corresponding to the inspection time points before and after is formed, the time-series twin mapping module continuously performs damage area expansion comparison processing on the damage boundary, damage coverage area and damage expansion direction in the inspection time points before and after, so that the boundary relationship, coverage relationship and expansion direction relationship in the continuously changing section are maintained, and the boundary relationship, coverage relationship and expansion direction relationship in the abnormally changing section are re-corresponded. Then, the corresponding boundary extension correspondence, coverage correspondence and expansion direction correspondence are written into the corresponding surface area in the hybrid tower digital twin data to generate damage time-series mapping data. After receiving the damage time-series mapping data, the early warning output module sequentially extracts the regional change content, the tower body part change content, and the damage expansion content. It then performs regional change rate analysis on the regional change content, continuous tower body part change analysis on the tower body part change content, and damage expansion direction analysis on the damage expansion content, ensuring that the change amplitude relationship, part continuity relationship, and direction change relationship under the same spatial location remain synchronously corresponding. Subsequently, it performs early warning result generation processing on the regional change rate result, the continuous tower body part change result, and the damage expansion direction result, forming damage evolution early warning data and hybrid tower digital twin visualization early warning results.

[0077] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital twin-based intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures, characterized in that: include: The inspection image acquisition module performs time stamp association processing, spatial location association processing, and part correspondence processing on images of the tower base area, concrete tower section, steel-concrete connection section, upper tower section, and shooting posture information to generate mixed tower inspection image data. The perspective unification processing module performs image distortion correction, image scale unification, image perspective registration, and tower surface correlation processing on mixed tower inspection image data to generate perspective unified image data. The 3D twin modeling module performs multi-view reconstruction processing, surface point cloud generation processing, tower surface mesh fitting processing, and texture mapping processing on image data with unified perspective to generate hybrid tower 3D surface model data, and generates hybrid tower digital twin data based on the hybrid tower 3D surface model data. The damage image analysis module performs crack region segmentation, peeling region segmentation, corrosion region segmentation, and damage edge contour refinement on the image data with a unified viewpoint to generate damage segmentation result data. The damage 3D projection module performs damage region pose projection processing, surface landing point correspondence processing, homogeneous damage spatial merging processing, and duplicate mapping reduction processing on the damage segmentation result data and the mixed tower 3D surface model data to generate damage 3D projection data. The temporal twin mapping module performs time alignment processing, spatial location inheritance processing, damage region expansion comparison processing, and twin surface evolution writing processing on the three-dimensional projection data of damage corresponding to different inspection time points to generate temporal damage mapping data. The early warning output module performs regional change rate analysis, continuous change analysis of tower body parts, damage expansion direction analysis, and early warning result generation on the damage time-series mapping data, and outputs damage evolution early warning data and hybrid tower digital twin visualization early warning results.

2. The intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures based on digital twins as described in claim 1, characterized in that: The inspection image acquisition module receives images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section, as well as the shooting pose information corresponding to each image, collected at different inspection times of the hybrid tower structure. Images of the tower base area, concrete tower section, steel-concrete connection section, and upper tower section are time-stamped and associated according to the inspection time point. Perform spatial location association processing on the captured pose information and the corresponding image; The image content is processed to correspond with the structural parts of the tower; The results of time stamp association processing, spatial location association processing, and part correspondence processing are uniformly integrated to generate mixed tower inspection image data.

3. The intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures based on digital twins as described in claim 1, characterized in that: The unified view processing module receives mixed tower inspection image data and extracts images of the tower base area, concrete tower section, steel-concrete connection section, upper tower section, and shooting pose information from the mixed tower inspection image data. Image distortion correction and image scale unification processing were performed on the images of the tower base area, the concrete tower section, the steel-concrete connection section, and the upper tower section. Based on the shooting pose information, perform image viewpoint registration on the image after image distortion correction and image scale unification processing; Perform correlation processing on the tower surface of the image after image viewpoint registration; After completing the correlation processing of the tower surface, correlation writing processing is performed to generate image data with unified perspective.

4. The intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures based on digital twins as described in claim 1, characterized in that: The 3D twin modeling module receives viewpoint-unified image data and extracts image location content, image texture content, and surface unfolding location content from the viewpoint-unified image data. Perform multi-view reconstruction processing on the image location content, image texture content, and surface unfolded location content to generate the corresponding reconstruction results; The reconstructed results are processed by generating surface point clouds and fitting tower surface meshes. After completing the mesh fitting process on the tower surface, perform texture mapping processing on the image texture content to generate hybrid tower 3D surface model data. Perform twin surface correspondence writing processing on the hybrid tower 3D surface model data to generate hybrid tower digital twin data.

5. The intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures based on digital twins as described in claim 1, characterized in that: The damage image analysis module receives unified view image data and mixed tower 3D surface model data. It extracts the initial damage seed regions from the crack area, spalling area and corrosion area and adjacent view images from the unified view image data. It uses a seed region growth algorithm and combines the boundary correspondence in adjacent view images, the corresponding surface position in the mixed tower 3D surface model data and the corresponding positional relationship of the tower base area, concrete tower section, steel-concrete connection section and upper tower section to perform seed retention processing on each initial damage seed region, so that each initial damage seed region forms a retained damage seed region within the corresponding part range. The boundary expansion process is performed on the preserved damage seed region. Adjacent pixels that are consistent with the current damage seed region in terms of grayscale variation, texture direction, boundary continuity and location continuity are incorporated into the current damage seed region. Adjacent pixels that are not continuous with the current damage seed region in terms of texture direction deviation and corresponding surface position offset are excluded from the boundary expansion range, thus forming the expanded damage region boundary. Feedback write-back processing is performed on the expanded damage region boundary. Based on the alignment results between the expanded damage region boundary and the corresponding boundary in the adjacent view image, the fitting results between the expanded damage region boundary and the corresponding surface position in the mixed tower 3D surface model data, and the continuity results between the boundary boundary and the corresponding positional relationship, the continuous and stable boundary direction is further expanded, and the boundary direction with continuous offset and discontinuous positional relationship is backed up. Finally, the damage region boundary after the completion of the further expansion and boundary backing is updated to form the damage segmentation result data.

6. The intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures based on digital twins as described in claim 1, characterized in that: The damage 3D projection module receives damage segmentation result data and mixed tower 3D surface model data. It extracts damage boundary segments from images from different perspectives from the damage segmentation result data. It uses a seed region growing algorithm to perform damage region pose projection processing on the damage segmentation result data and mixed tower 3D surface model data. Based on the surface position relationship of the tower base area, concrete tower segment, steel-concrete connection segment and upper tower segment, it performs trimming correspondence processing on the damage boundary segments projected to the same location and with adjacent projection positions. It removes correspondence relationships that are inconsistent with the current projection position, boundary direction and location relationship, forming the damage boundary segments after trimming correspondence processing. Point-to-surface convergence processing is performed on the damaged boundary segments after trimming and the local surface segments in the three-dimensional surface model data of the mixed tower. Based on the fitting results of each damaged boundary segment to the local surface normal direction, the positional continuity relationship between local surface segments, and the surface connection relationship within the corresponding part range, the projection position of each damaged boundary segment in the three-dimensional surface model data of the mixed tower is updated to form the updated damaged boundary segments. The updated damage boundary segments are subjected to spatial merging processing of the same source damage. Damage boundary segments that are continuous and consistent in spatial location, boundary orientation, surface fitting results and location relationship are merged into the same damage region. The merged damage region is then subjected to repeated mapping reduction processing, and a single projection region corresponding to the damage region is retained in the mixed tower 3D surface model data to form 3D damage projection data.

7. The intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures based on digital twins as described in claim 1, characterized in that: The temporal twin mapping module receives damage 3D projection data and mixed tower digital twin data; Time alignment processing is performed on the 3D projection data of damage corresponding to different inspection time points according to the inspection time point to generate time alignment results; The time alignment results and the hybrid digital twin data are processed according to their spatial location to generate spatial inheritance results. A dynamic time warping algorithm combined with a Hampel filtering algorithm is used to perform damage region expansion comparison processing on spatial inheritance results, time alignment results and mixed tower digital twin data to generate expansion comparison results. The extended alignment results are processed by twin surface evolution writing to generate damage time-series mapping data.

8. The intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures based on digital twins as described in claim 7, characterized in that: The dynamic time warping algorithm performs damage region expansion comparison processing on the spatial inheritance results, time alignment results, and mixed tower digital twin data, including the following steps: The three-dimensional projection data of damage corresponding to the previous inspection time point and the three-dimensional projection data of damage corresponding to the current inspection time point in the spatial inheritance result are processed by continuous region correspondence according to spatial position relationship. Combined with the time alignment result, the damage boundary, damage coverage area and damage expansion direction in the previous inspection time point and the damage boundary, damage coverage area and damage expansion direction in the current inspection time point are extracted to form the original damage change sequence corresponding to the previous inspection time point and the original damage change sequence corresponding to the current inspection time point. The Hampel filtering algorithm is used to perform sliding window anomaly identification and anomaly replacement processing on the original damage change sequence corresponding to the previous inspection time point and the original damage change sequence corresponding to the current inspection time point, respectively. Segmentation and aggregation processing is performed on continuous anomaly positions to form the smooth damage change sequence corresponding to the previous inspection time point, the smooth damage change sequence corresponding to the current inspection time point, the anomaly segmentation result corresponding to the previous inspection time point, and the anomaly segmentation result corresponding to the current inspection time point. Perform position-by-position local distance calculation on the smooth damage change sequence corresponding to the previous inspection time point and the smooth damage change sequence corresponding to the current inspection time point to generate a basic distance matrix. Then, combine the abnormal segmentation results corresponding to the previous inspection time point and the abnormal segmentation results corresponding to the current inspection time point to perform abnormal position correction processing on the basic distance matrix to generate a corrected distance matrix. The modified distance matrix is ​​combined with the abnormal segment results corresponding to the previous inspection time point and the abnormal segment results corresponding to the current inspection time point to perform regular window self-shrinking processing. The width of the regular window is reduced for continuously changing segments, and the local elastic regular window is retained for the segments corresponding to abnormal segments to generate regular window results. Within the normalized window result, step cost renormalization is performed on the modified distance matrix. The cumulative cost is increased for matching units within abnormal segments and matching units that cross the boundaries of abnormal segments, while the basic cost is maintained for matching units within continuously changing segments, thus generating the step renormalization result. Perform forward regularization path search and reverse regularization path search on the step reorganization results respectively to generate forward regularization path results and reverse regularization path results; Perform bidirectional path consistency verification on the forward regularization path results and the reverse regularization path results. Determine the segments with consistent or adjacent path mapping relationships as stable alignment segments, and determine the segments with deviated path mapping relationships and corresponding abnormal segments as unstable alignment segments, and generate path consistency results. For unstable alignment segments in the path consistency results, combine the abnormal segment results corresponding to the previous inspection time point, the abnormal segment results corresponding to the current inspection time point, and the corresponding surface areas in the mixed tower digital twin data to perform write-back and rollback processing. Then, within the local regularization window after rollback, perform dynamic time regularization path search processing again to generate write-back and re-regularization results. The stable alignment section and the write-back re-normalization results are uniformly integrated and processed. In combination with the corresponding surface areas in the hybrid tower digital twin data, boundary extension correspondence processing, coverage correspondence processing, expansion direction correspondence processing and position consistency verification processing are performed to generate extended comparison results.

9. The intelligent sensing and early warning system for full-life-cycle damage of hybrid tower structures based on digital twins as described in claim 1, characterized in that: The early warning output module receives damage time-series mapping data and extracts regional change content, tower body part change content, and damage expansion content from the damage time-series mapping data. Perform regional change rate analysis on the changes in the region and generate regional change rate results; Perform continuous change analysis on the changes in the tower body parts to generate continuous change results for the tower body parts; Perform damage propagation direction analysis on the damage propagation content to generate damage propagation direction results; The system performs early warning result generation processing on the regional change rate results, the continuous change results of tower body parts, and the damage propagation direction results, and outputs damage evolution early warning data and mixed tower digital twin visualization early warning results.

Images