Space structure health monitoring method and system based on sky-ground cooperation

Abstract

The invention discloses a space structure health monitoring method and system based on sky-ground cooperation, and relates to the technical field of space structure health monitoring. The method comprises the following specific steps: respectively monitoring a space structure in three dimensions of sky, air and ground to obtain multi-dimensional monitoring information; performing data space-time synchronization on the multi-dimensional monitoring information to obtain panoramic data; performing comprehensive evaluation and life evolution trend prediction on static and dynamic performance of the space structure based on foundation layer sensing data; performing multi-source data feature extraction and fusion on the panoramic data, constructing a comprehensive risk index evaluation model, and performing three-level graded early warning; after early warning is triggered, an LLM model is introduced, and a disposal scheme is automatically generated according to an early warning structure. According to the invention, three monitoring means of satellite remote sensing, unmanned aerial vehicle inspection and a ground sensing network are deeply fused, so that all-directional, multi-scale and multi-dimensional sensing and monitoring of the space structure in the whole period of construction and operation are realized.

Description

Technical Field

[0001] This invention relates to the field of space structure health monitoring technology, and in particular to a space structure health monitoring method and system based on sky-ground coordination. Background Technology

[0002] In recent years, with the continuous expansion of urban construction, large-span stadiums, airport terminals, convention centers, and other large spatial structures have been widely used both domestically and internationally. Their safe operation and life-cycle management have increasingly attracted high attention from all sectors of society and academia. However, these structures generally possess characteristics such as large spans, complex component systems, unique stress mechanisms, and susceptibility to significant external environmental influences, resulting in two major technical shortcomings in existing monitoring methods: First, existing monitoring methods largely rely on sensors to collect local responses from key structural components, making it difficult to comprehensively and accurately reflect the overall safety status of the structure. Although some research has attempted to infer overall structural performance from local response data, in practical engineering applications, the superposition of multiple factors such as temperature fluctuations, random load changes, and environmental noise interference can easily lead to deviations in the accuracy of performance inferences, resulting in distorted structural safety assessments and failing to provide a reliable basis for operation and maintenance decisions. Second, most existing monitoring systems remain at the initial stage of "data perception and acquisition," lacking intelligent analysis of monitoring data, structural risk prediction, and operation and maintenance decision support functions, and have not yet formed a complete intelligent management closed loop of "perception-analysis-prediction-decision." When potential safety hazards appear in the structure, the system is unable to provide early warnings and can only passively respond after a fault occurs, failing to meet the needs of safety management throughout the entire life cycle of large spatial structures. Therefore, for those skilled in the art, the current field of spatial structural health monitoring urgently needs to establish a multi-source perception system that coordinates space, air, and ground, and to achieve accurate structural status assessment, early risk warning, and a closed-loop management system throughout the entire chain through intelligent technologies. Summary of the Invention

[0003] The purpose of this invention is to provide a space structure health monitoring method and system based on sky-ground collaboration to solve the problems mentioned in the background technology. It deeply integrates three types of monitoring methods: satellite remote sensing, UAV inspection, and ground sensor network, to achieve all-round, multi-scale, and multi-dimensional perception and monitoring of space structures throughout the entire construction and operation cycle.

[0004] To achieve the above objectives, the present invention provides the following solution: On one hand, it provides a method for monitoring the health of space structures based on sky-ground coordination, the specific steps of which include the following: The spatial structure is monitored separately in three dimensions: sky, air, and ground, to obtain multi-dimensional monitoring information; the multi-dimensional monitoring information is then spatiotemporally synchronized to obtain panoramic data. Comprehensive evaluation of the static and dynamic performance of spatial structures and prediction of their life evolution trends based on ground-level sensing data; Multi-source data feature extraction and fusion are performed on the panoramic data to construct a comprehensive risk index assessment model and conduct three-level graded early warning. After an early warning is triggered, an LLM model is introduced to automatically generate a response plan based on the early warning structure.

[0005] Preferably, the specific steps for monitoring the spatial structure at the ground level are as follows: Multi-temporal high-resolution images were acquired using BeiDou satellite positioning and synthetic aperture radar interferometry, and image preprocessing was completed to obtain the input images; The input image is used to identify buildings and extract bounding boxes using a deep learning object detection algorithm. Differential analysis is performed on the detection results of the same area at different times to obtain the potential range of building changes. The potential range of change of the building is further identified by a pixel-level Softmax classification layer, and a change probability map is output. The change probability map is subjected to morphological post-processing and noise suppression operations to obtain a change mask, which is superimposed on the building vector boundary around the spatial structure to form a dynamically updated environmental evolution database.

[0006] Preferably, monitoring the spatial structure at the ground level also includes acquiring regional atmospheric environmental monitoring data, including temperature, humidity, air pressure, wind speed, wind direction, and rainfall. The regional atmospheric environmental monitoring data is combined with a disaster prediction model to sense extreme weather events. When the predicted indicators exceed a set threshold, high-frequency sampling of ground sensors and intensified inspection by UAVs in the airspace are automatically triggered to achieve dynamic adjustment of the monitoring strategy.

[0007] Preferably, the monitoring of the spatial structure is achieved by using drones equipped with multiple sensors on the air base layer, including the identification of apparent defects and the monitoring of overall deformation.

[0008] Preferably, the specific method for monitoring the overall deformation of the air-base layer is as follows: A UAV equipped with a lidar acquires high-density three-dimensional point cloud data; a high-precision three-dimensional geometric model of the spatial structure is generated through point cloud filtering, registration, and modeling algorithms; based on point cloud differential analysis, the deformation at any given time is calculated. t Compared with reference state t The geometric offset of 0 enables the identification of changes in the overall arch of the roof and the displacement of key components. At the same time, a dynamic database of the evolution of the structural morphology over time is established through multi-temporal point cloud registration.

[0009] Preferably, the specific method for identifying the apparent defects in the empty substrate is as follows: The YOLO series target detection network is used to analyze high-definition images collected by UAVs. During a forward propagation, the category labels of apparent diseases and their rectangular bounding boxes are output to achieve the initial screening of diseases in large-area images. The candidate regions output by YOLO are used as regions of interest and input into a pixel-level segmentation network to perform fine segmentation of the disease boundaries and generate pixel-level masks. The pixel-level mask is used to quantify various surface defects.

[0010] Preferably, a multi-dimensional sensing system is formed at the ground level, supported by DIC displacement monitoring, MEMS multi-parameter environmental and dynamic sensing, and independent mechanical monitoring, and unified access is achieved through a wireless network.

[0011] Preferably, the steps for constructing the comprehensive risk index assessment model and performing three-level graded early warning are as follows: Environmental threat level E Overall health H With local security S The indicators from the three dimensions are uniformly transformed into standardized features to construct the comprehensive risk index assessment model: ; in, α , β , γ The weighting coefficients are determined by expert weighting or the analytic hierarchy process, satisfying 𝛼+𝛽+𝛾=1; Based on meteorological and remote sensing image data acquired from surface-level monitoring, external load parameters are extracted, and an environmental risk index is calculated using a normalization method to obtain the environmental threat level. E The overall health score is obtained by forming a global health index based on point cloud data acquired from the spatial layer monitoring through fuzzy comprehensive evaluation. H Based on the number of times local monitoring parameters exceeded limits, cable imbalance coefficient, and cumulative fatigue damage obtained from ground level monitoring, the local safety margin of key components is evaluated to obtain the local safety margin. S ; Comprehensive risk index R The results are compared with the tiered thresholds to generate multi-level early warning results.

[0012] On the other hand, a space structure health monitoring system based on sky-ground collaboration is provided, including a multi-dimensional monitoring module, a data processing module, a performance evaluation module, a risk early warning module, and an intelligent decision-making module; among which, The multi-dimensional monitoring module is used to monitor the spatial structure in the three dimensions of sky, air, and ground respectively, and obtain multi-dimensional monitoring information. The data processing module is used to perform spatiotemporal synchronization of the multi-dimensional monitoring information to obtain panoramic data; The performance evaluation module is used to comprehensively evaluate the static and dynamic performance of spatial structures and predict their life evolution trends based on the ground-level sensing data structure. The risk warning module is used to extract and fuse multi-source data features from the panoramic data, construct a comprehensive risk index assessment model, and conduct three-level graded warnings. The intelligent decision-making module is used to introduce an LLM model after an early warning is triggered and automatically generate a response plan based on the early warning structure.

[0013] Preferably, the multi-dimensional monitoring module, the data processing module, and the performance evaluation module are integrated. The risk warning module and the intelligent decision-making module are integrated through a lightweight integrated platform and connected to a multi-terminal collaborative management module.

[0014] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: Through a three-layered collaborative sensing system encompassing the air, space, and ground, comprehensive coverage is achieved, encompassing macroscopic environmental disturbances, overall geometric morphology, and local mechanical responses. In the air layer, InSAR and satellite remote sensing are used to rapidly acquire large-scale morphological evolution data. In the ground layer, UAV point clouds and target detection enable mesoscale overall deformation and defect identification. At ground level, DIC and MEMS sensor networks are used to collect local displacement, strain, and cable force data, thus avoiding the limitations of single-method approaches and significantly improving monitoring efficiency. A three-tiered risk assessment system based on environmental threat level, overall health level, and local security level was proposed. Through multi-source data fusion, a comprehensive risk index was established, which can achieve graded early warning at different levels. This mechanism not only improves the sensitivity of early warning, but also enhances the foresight and interpretability of early warning. By introducing AI and large language models into the platform application layer, a response plan can be automatically generated after an early warning is triggered. Combined with secondary review by experts, this forms an "AI + human" collaborative decision-making model. This mechanism significantly improves the efficiency from risk identification to emergency response, enabling the system to move from "passive perception" to "proactive management." Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1This is the overall technical roadmap of the present invention; Figure 2 A roadmap for grassroots monitoring technology; Figure 3 A technical roadmap for airborne grassroots monitoring; Figure 4 A technical roadmap for grassroots monitoring; Figure 5 This is a technical roadmap for structural performance assessment and prediction based on ground-level data; Figure 6 A technical roadmap for risk warning and intelligent decision-making based on multi-source data fusion. Detailed Implementation

[0017] 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.

[0018] The purpose of this invention is to provide a space structure health monitoring method based on sky-ground coordination, the overall technical roadmap is as follows: Figure 1 As shown, the specific steps include the following: S1. Monitor the spatial structure separately in the three dimensions of sky, air, and ground to obtain multi-dimensional monitoring information; synchronize the multi-dimensional monitoring information in time and space to obtain panoramic data; S2. Based on the ground-level sensing data structure, comprehensively evaluate the static and dynamic performance of spatial structures and predict their life evolution trend; S3. Extract and fuse multi-source data features from panoramic data, construct a comprehensive risk index assessment model, and conduct three-level graded early warning. S4. After the warning is triggered, an LLM model is introduced to automatically generate a response plan based on the warning structure.

[0019] In step S1, multi-source data is collected, encompassing three data acquisition pathways: sky, air, and ground. Sky corresponds to environmental loads and surrounding building dynamics; air corresponds to external defects and overall deformation; and ground corresponds to deformation and strain of local structural components. These represent three levels of data. Step S2, processing the ground-level data, has the highest priority among these three levels, as it most directly reflects real-time structural changes. Therefore, the structural monitoring data acquired in step two first needs to be used for spatial structural performance assessment and lifespan evolution prediction to directly reflect the structural state. In step S3, the three types of data are fused and learned to form a panoramic data matrix of "external environment - overall state - local stress," avoiding data silos. Compared to other methods that only use step two, this invention analyzes data from multiple dimensions, ultimately enhancing the accuracy of early warnings.

[0020] Furthermore, in S1, multi-dimensional monitoring information perception and data synchronization are carried out. This step mainly involves collecting, calibrating, and registering the environmental, morphological, and mechanical response data of the "sky-air-ground" three layers into the same spatiotemporal reference, forming a high-quality, computable panoramic data base, and providing reliable input for subsequent performance evaluation, trend prediction, and risk warning.

[0021] Tianchi monitoring like Figure 2 The diagram shows the monitoring route at the ground level. At the ground level, the system utilizes BeiDou satellite positioning and Synthetic Aperture Radar Interferometry (InSAR) technology to acquire multi-temporal high-resolution images. Image preprocessing, including orthorectification and radiometric consistency, ensures the comparability of data from different times across spatial locations and lighting conditions. Subsequently, deep learning target detection algorithms (YOLO series models) are used to quickly identify buildings and extract bounding boxes from the images. Combined with a change detection process, differential analysis is performed on the detection results of the same area at different times to determine the extent of newly added, modified, or demolished buildings. ; in, B t and B t−1 They represent the times at time 1 and 2 respectively. t and t -1 is the set of building boundaries, Δ B It represents the dynamic changes in the surrounding environment. Furthermore, by combining the classification results of a pixel-level convolutional neural network (CDNet), the system can accurately identify changes in ground features outside building boundaries.

[0022] Specifically, firstly, multi-temporal remote sensing images are registered and normalized to ensure spatial and spectral consistency of the input images; then, a two-stream convolutional neural network is constructed to process the temporal data. t With time tThe images at -1 are input into the feature extraction subnetwork to obtain high-dimensional feature tensors. F t and F t−1 In the change discrimination stage, the difference operation Δ is used. F =∣ F t - F t−1 | Alternatively, a feature fusion mechanism is used to compare features from two time periods and extract potential change areas. Finally, a pixel-level Softmax classification layer is used to divide each pixel into "unchanged" and "changed" categories, and a change probability map is output. After morphological post-processing and noise suppression, a clear change mask is obtained, which is then superimposed on the vector boundaries of buildings surrounding the spatial structure to form a dynamically updated environmental evolution database. The advantage of this method is that, compared with traditional threshold-based difference methods, convolutional neural networks can capture more complex spectral and spatial feature differences, effectively reduce spurious changes caused by illumination, shadows, clouds, and fog, and significantly improve the accuracy of identifying changes in buildings and surrounding features.

[0023] Meanwhile, monitoring spatial structures at the stratum level also includes access to regional atmospheric environmental monitoring and meteorological numerical models, encompassing multi-dimensional environmental parameters such as temperature, humidity, air pressure, wind speed, wind direction, and rainfall. This is combined with disaster prediction models to provide early detection of extreme events such as typhoons, rainstorms, and strong winds. When predicted indicators exceed set thresholds, high-frequency sampling by ground sensors and intensive UAV patrols of the stratum are automatically triggered, enabling dynamic adjustments to the monitoring strategy. Through a process of "remote sensing image change detection—building update identification—environmental disturbance prediction—monitoring strategy linkage," the stratum level not only provides macroscopic evolutionary information on spatial structures and their surrounding environment but also establishes a proactive early warning mechanism for hazardous disturbances. This provides a global benchmark and external boundary conditions for overall stratum deformation monitoring and local stratum mechanical data acquisition. In the entire air-space-ground collaborative system, the stratum level acts as both a global benchmark provider and a proactive risk perceiver, providing a macroscopic framework and boundary conditions for subsequent local and detailed monitoring of the stratum and ground.

[0024] Empty base monitoring like Figure 3 The diagram shows the monitoring roadmap for the air-base layer, which primarily relies on UAV platforms equipped with multiple sensors to achieve mesoscale monitoring of the spatial structure. Monitoring of the spatial structure within the air-base layer is achieved using UAVs equipped with multiple sensors, including identification of apparent defects and monitoring of overall deformation.

[0025] The overall deformation monitoring method is as follows: by calculating the overall displacement field of the structure (such as roof deformation, support horizontal displacement, and truss displacement), the long-term deformation trend of the structure is quantitatively assessed; combined with the structural design limits, the overall deformation exceeding the safe range is identified in a timely manner, providing macroscopic data support for judging the overall stability of the structure.

[0026] The specific method for overall deformation monitoring is as follows: A UAV equipped with a LiDAR (Light Detection and Ranging) system acquires high-density 3D point cloud data. Point cloud filtering, registration, and modeling algorithms are then used to generate a high-precision 3D geometric model of the structure. Based on point cloud differential analysis, the model can be calculated at any given time. t Compared with reference state t Geometric offset of 0: ; This enables the identification of changes in the overall arch of the roof and the displacement of key components. At the same time, through multi-temporal point cloud registration, a dynamic database of the structural morphology evolving over time is established, providing a basis for the geometric calibration of the digital twin model.

[0027] Furthermore, the identification of apparent defects in the bare base layer adopts a cascaded detection process using YOLO+Mask, specifically as follows: First, the YOLO series object detection network is used to quickly analyze high-definition images acquired by UAVs. During a single forward propagation, it outputs category labels and rectangular bounding boxes for defects such as cracks, corrosion, leakage, and node fatigue, achieving efficient initial screening of defects in large-area images. Then, the candidate regions output by YOLO are input as regions of interest (ROIs) into a pixel-level segmentation network (Mask R-CNN) to finely segment the defect boundaries, generating pixel-level masks. Based on the mask results, various quantifications can be achieved: crack length is calculated by accumulating the distances of each segment after skeleton extraction; corrosion or leakage area is converted into the actual area by statistically analyzing the total number of mask pixels; and the defect expansion rate is calculated using multi-temporal area difference Δ. A / Δ t This process combines the rapid detection capabilities of YOLO with the fine quantification capabilities of Mask networks, thereby achieving efficient identification and accurate quantification of surface defects.

[0028] It is evident that the data fusion of the empty base layer provides dual information on the overall geometric morphology of the structure and the evolution of apparent defects. On the one hand, the overall deformation data of the point cloud model provides quantitative indicators for the macroscopic stability and morphological preservation of the structure; on the other hand, the defect features identified by target detection provide visual evidence for anomalies in formation strain and cable force data. Through the process of "point cloud modeling—geometric difference—defect detection—feature fusion," a mesoscale supplement and verification of the local mechanical data of the formation is formed.

[0029] Ground-level monitoring like Figure 4The diagram shows the technical roadmap for ground level monitoring. The focus is on acquiring local responses and environmental parameters of key components of the spatial structure at the ground level to build the most direct and fine-grained data foundation. First, digital image correlation (DIC) technology is used to deploy targets at locations such as mid-span and supports. A series of images are acquired using a multi-camera array, and the displacement field is calculated using sub-pixel grayscale correlation methods. ; Where C is the correlation coefficient. f , g These represent the reference image and the deformed image, respectively. This method achieves displacement measurement with millimeter to sub-millimeter accuracy without physical contact, making it suitable for scenarios where traditional displacement gauges are difficult to deploy in large-span spatial structures.

[0030] Secondly, the multi-parameter integrated sensing node developed based on MEMS technology can realize real-time acquisition of environmental and dynamic parameters such as acceleration, tilt angle, and temperature. It has the advantages of miniaturization, low power consumption and wireless capability, and is suitable for large-scale distributed deployment and long-term operation.

[0031] This invention's monitoring system enables precise local monitoring of various spatial structures. Monitoring parameters include essential basic parameters and optional adaptive parameters: essential basic parameters include displacement and stress, while optional adaptive parameters include cable force, support settlement, and membrane tension. Considering the differences in the structural form, stress characteristics, and engineering requirements of spatial structures, optional adaptive parameters are not mandatory for all spatial structures. For spatial structures with specific structural stress characteristics, corresponding adaptive monitoring modules can be activated simultaneously to achieve real-time monitoring of specific conditions. For spatial structures without such structural or stress characteristics, only essential basic monitoring items such as displacement and stress can be activated, and the corresponding adaptive monitoring modules can be omitted as needed. This invention forms a ground-based monitoring system characterized by "lightweight deployment, high-precision monitoring, and low-power operation and maintenance." It allows for flexible selection and module combination based on the actual characteristics of the spatial structure, possessing both versatility and specific adaptability, and meeting the monitoring needs of various spatial structures.

[0032] For example, for core structural mechanics parameters such as strain and cable force, more mature sensing technologies are used for independent monitoring: strain monitoring primarily utilizes fiber optic gratings (FBGs) or vibrating wire strain gauges to ensure long-term stability and high accuracy; cable force monitoring is achieved through surface strain methods or anchor pressure ring methods, and combined with calibration models, the stress state of the cables can be quantified. These key mechanical sensing units are also connected to a unified wireless sensor network, forming a complete ground monitoring system together with MEMS nodes.

[0033] To avoid the complexities and maintenance difficulties associated with cable routing in long-span structures, this invention employs a unified wireless sensor network for data acquisition and transmission. Millisecond-level synchronization is achieved through LoRa / 5G multi-hop self-organizing networks and BeiDou / GPS high-precision time synchronization. Nodes possess edge computing capabilities, enabling local feature extraction (such as RMS values, fatigue cycle counts, and spectral peak values), uploading only key features or abnormal data, significantly reducing energy consumption and communication burden. In summary, a multi-dimensional sensing system supported by DIC displacement monitoring, MEMS multi-parameter environmental and dynamic sensing, and independent mechanical monitoring is formed at the ground level, achieving unified access through a wireless network. This provides a direct and reliable data source for performance evaluation, trend prediction, and risk warning.

[0034] After completing their respective monitoring at the air, space, and ground levels, unified fusion is achieved through a multi-source data synchronization mechanism. Firstly, in the time dimension, all sensor nodes and remote sensing images are synchronized using BeiDou / GPS timing and 5G network clock correction, ensuring data from different sources are aligned within milliseconds, avoiding misjudgments of causal relationships due to time drift. Secondly, in the spatial dimension, the system uses the BIM 3D coordinate system as a unified benchmark to perform orthorectification on remote sensing images, registration on UAV LiDAR point clouds, and precise calibration of ground sensor deployment locations, thereby achieving spatial alignment of cross-layer data. Thirdly, at the data level, resampling, interpolation, and compressed sensing technologies are used to transform high-frequency dynamic monitoring data and low-frequency environmental monitoring data into a time-series data stream with a unified frequency and format. To eliminate environmental disturbances, all monitoring signals employ a unified decomposition model. ; in, F env ( t ) represents the environmental load effect. R struct ( t () represents the actual structural response. This represents the noise term. The model achieves "environmental effect stripping—net response extraction," ensuring the reliability of the multi-source data fusion. Ultimately, the system constructs a panoramic data foundation aligned in the spatiotemporal dimensions, unified in the numerical dimensions, and with clear physical meaning. This allows the macroscopic environmental benchmark of the celestial layer, the overall geometric deformation and damage information of the stratigraphic layer, and the local mechanical data of the geological strata to complement and verify each other. This synchronous fusion mechanism provides a solid foundation for subsequent structural performance assessment, trend prediction, and graded risk early warning.

[0035] It should be noted that the monitoring parameters collected by the monitoring system will vary depending on the spatial structure being monitored. After completing the multi-dimensional information synchronization, the system enters the performance evaluation and prediction stage based on ground-level sensing data, where the acquired ground-level data is analyzed, evaluated, and predicted accordingly. For example... Figure 5 As shown, this stage uses multi-parameter monitoring data such as displacement, strain, cable force, acceleration, temperature, humidity and wind pressure as input. After signal processing and theoretical modeling, it realizes the comprehensive evaluation of the static and dynamic performance of the structure and the prediction of its life evolution trend.

[0036] Firstly, in terms of performance evaluation, load-response separation and modal identification methods are used to process the monitoring signals. Through an improved ensemble empirical mode decomposition and independent component analysis (EEMD-ICA), the monitoring sequence is decomposed into environmental load effects. F env ( t ) and net response R struct ( t The local "net response" of the structure under "pure load" is extracted to eliminate non-load interference for subsequent performance evaluation and improve the accuracy of the evaluation. Based on the local "net response" data after removing environmental interference, a targeted local performance evaluation system is constructed from three dimensions: static, dynamic, and durability. The static performance evaluation focuses on key local components, using indicators such as peak strain, cable force deviation rate, and displacement exceedance number to determine whether the local stress exceeds the design limit; the dynamic performance evaluation combines MEMS sensing data to analyze the local stiffness degradation; the durability evaluation, based on long-term monitored strain cycle number and environmental humidity data, calculates the cumulative fatigue damage of local components and determines their remaining service life; through multi-dimensional weighted scoring, a local performance level is formed, which intuitively reflects the health status of key parts of the structure.

[0037] Secondly, in terms of performance prediction, a hybrid strategy of "mechanism model - signal decomposition - deep learning prediction" is adopted to improve the accuracy and robustness of the prediction. First, the ground-level monitoring data is preprocessed, and the original signal is decomposed into multi-scale intrinsic mode functions (IMFs) and low-frequency residuals using ensemble empirical mode decomposition (EEMD). ; Among them, the high-frequency IMF reflects environmental noise and short-term disturbances, while the low-frequency IMF reflects residuals. r ( t This better reflects the structural degradation trend over time. By retaining the low-frequency net response, the interference of short-term fluctuations on the prediction model is avoided.

[0038] In the data-driven prediction phase, a bidirectional long short-term memory (Bi-LSTM) network is used to train and predict the net response sequence. Compared to traditional LSTM, Bi-LSTM can simultaneously capture the forward and backward dependencies of the time series, making it more suitable for complex time series such as structural responses, which exhibit "cumulative and hysteresis effects." The prediction process can be represented as follows: ; in, This is a Bi-LSTM prediction function, with low-frequency components and residuals as inputs, and the predicted response at the next time step as the output. .

[0039] To enhance the reliability of the predictions, a Bayesian Dynamic Linear Model (BDLM) is further introduced as a mechanism correction module to correct the prediction results of Bi-LSTM. This is achieved through the state transition equation: ; in, x t For structural state, y t For observation purposes, A , C For the state transition and observation matrix, w t , v t This is a noise term. Through analysis of... x t Dynamic estimation and updating enable real-time prediction of structural states. Bayesian updates are applied to the prediction results to correct systematic biases in the deep learning model, ensuring that the prediction trend aligns with the laws of structural mechanics. Combined with long-term trend analysis, if the prediction results show that response indicators consistently approach or exceed design thresholds, the system will automatically trigger a tiered early warning mechanism.

[0040] Furthermore, multi-source data fusion risk warning is implemented in S3, with the following methodology and process: Figure 6 As shown, firstly, feature extraction and fusion are performed on monitoring data from the sky, air, and ground. Sky data focuses on the macroscopic features of environmental disturbances and surrounding changes; air data provides the geometric features of the overall structural morphology and apparent defects; and ground data reflects the mechanical features of local strain, cable force, and vibration. Through multi-source feature engineering and data alignment, a unified feature vector containing temporal, spatial, and physical attributes is constructed to achieve deep fusion of cross-layer data, ensuring that environmental effects, overall deformation, and local responses can be interpreted synergistically.

[0041] In the risk warning stage, the level of environmental threat will be... E Overall health H With local security SThe indicators from the three dimensions are uniformly transformed into standardized features to construct the comprehensive risk index assessment model: ; in, α , β , γ The weighting coefficients are determined by expert weighting or the analytic hierarchy process, satisfying 𝛼+𝛽+𝛾=1; Environmental threat level E Based on meteorological monitoring and remote sensing data, external load parameters such as typhoon wind speed, rainfall, and seismic intensity are extracted, and the environmental risk index is obtained by normalization method.

[0042] Overall health H Based on indicators such as the overall deformation of the point cloud, the rate of change of its natural frequency, and the rate of disease expansion, a global health index is formed through fuzzy comprehensive evaluation. ; Where, Δ d Δ represents the overall displacement change. ω The natural frequency drift rate, v damage This represents the rate of disease spread.

[0043] Local security S Combining the number of times local monitoring parameters exceeded limits, cable imbalance coefficient, and cumulative fatigue damage. D The local safety margin of key components is evaluated.

[0044] Comprehensive risk index R The results are compared with the tiered thresholds to generate multi-level early warning results: when R When at a low level, the system maintains a monitoring state; when R When the threshold is approached, a yellow alert is triggered and local retesting is recommended; when R When the threshold for a high level is exceeded, a red alert is triggered and an emergency response plan is automatically generated.

[0045] In the intelligent decision-making process, artificial intelligence and Large Language Modeling (LLM) are introduced to transform early warning results into actionable management and response plans. When an early warning at a certain level is triggered, the AI ​​engine first generates preliminary response suggestions based on monitoring data, including temporary control measures, retesting strategies, and long-term reinforcement plans. Subsequently, the Large Language Model generates structured intelligent decision text through interactive retrieval of monitoring databases, historical case libraries, and regulatory provisions, and supports multi-terminal push notifications. Through the "AI initial screening + expert secondary review" model, efficient and reliable risk management is achieved.

[0046] On the other hand, a space structure health monitoring system based on sky-ground collaboration is provided, including a multi-dimensional monitoring module, a data processing module, a performance evaluation module, a risk early warning module, and an intelligent decision-making module; among which, The multi-dimensional monitoring module is used to monitor the spatial structure separately in the three dimensions of sky, air, and ground to obtain multi-dimensional monitoring information; The data processing module is used to perform spatiotemporal synchronization of the multi-dimensional monitoring information to obtain panoramic data; The performance evaluation module is used to comprehensively evaluate the static and dynamic performance of spatial structures and predict their life evolution trends based on the ground-level sensing data structure. The risk warning module is used to extract and fuse multi-source data features from the panoramic data, construct a comprehensive risk index assessment model, and conduct three-level graded warnings. The intelligent decision-making module is used to introduce an LLM model after an early warning is triggered, and automatically generate a response plan based on the early warning structure.

[0047] Furthermore, the multi-dimensional monitoring module, data processing module, performance evaluation module, risk warning module, and intelligent decision-making module are integrated through a lightweight, unified platform and connected to a multi-terminal collaborative management module to form a closed-loop management model of "monitoring—evaluation—early warning—response—feedback". The platform adopts a cloud-edge collaborative and microservice architecture, supporting large-scale sensor access, real-time processing of multi-source data, and visualization. The platform has a built-in digital twin engine that constructs a virtual model of the spatial structure based on BIM and UAV point clouds, and realizes dynamic mapping between monitoring data and virtual objects. Through a 3D visualization interface, managers can intuitively view the overall deformation, local strain, and disease evolution of the structure.

[0048] The multi-terminal collaborative management module supports collaborative operation between PCs, mobile devices, and a cockpit-style large screen. The PC terminal focuses on historical data analysis, report generation, and policy configuration; the mobile terminal provides real-time alarm push notifications, on-site inspection feedback, and intelligent Q&A interaction; and the cockpit-style large screen emphasizes macro-level operational status display and emergency response management. All terminals connect through a unified interface to achieve information sharing and command interoperability, meeting the needs of multiple scenarios from daily monitoring to emergency response.

[0049] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for monitoring the health of space structures based on sky-ground coordination, characterized in that, The specific steps include the following: The spatial structure is monitored separately in three dimensions: sky, air, and ground, to obtain multi-dimensional monitoring information; the multi-dimensional monitoring information is then spatiotemporally synchronized to obtain panoramic data. Comprehensive evaluation of the static and dynamic performance of spatial structures and prediction of their life evolution trends based on ground-level sensing data; Multi-source data feature extraction and fusion are performed on the panoramic data to construct a comprehensive risk index assessment model and conduct three-level graded early warning. After an early warning is triggered, an LLM model is introduced to automatically generate a response plan based on the early warning structure.

2. The method for monitoring the health of space structures based on sky-ground coordination according to claim 1, characterized in that, The specific steps for monitoring spatial structures at the ground level are as follows: Multi-temporal high-resolution images were acquired using BeiDou satellite positioning and synthetic aperture radar interferometry, and image preprocessing was completed to obtain the input images; The input image is used to identify buildings and extract bounding boxes using a deep learning object detection algorithm. Differential analysis is performed on the detection results of the same area at different times to obtain the potential range of building changes. The potential range of change of the building is further identified by a pixel-level Softmax classification layer, and a change probability map is output. The change probability map is subjected to morphological post-processing and noise suppression operations to obtain a change mask, which is superimposed on the building vector boundary around the spatial structure to form a dynamically updated environmental evolution database.

3. The method for monitoring the health of space structures based on sky-ground coordination according to claim 1, characterized in that, Monitoring the spatial structure at the ground level also includes acquiring regional atmospheric environmental monitoring data, including temperature, humidity, air pressure, wind speed, wind direction, and rainfall. This regional atmospheric environmental monitoring data is combined with disaster prediction models to sense extreme weather events. When the predicted indicators exceed the set threshold, high-frequency sampling of ground sensors and intensified inspections by aerial drones are automatically triggered to achieve dynamic adjustment of the monitoring strategy.

4. The method for monitoring the health of space structures based on sky-ground coordination according to claim 1, characterized in that, In the air-base layer, drones equipped with multiple sensors are used to monitor the spatial structure, including the identification of apparent defects and the monitoring of overall deformation.

5. The method for monitoring the health of space structures based on sky-ground coordination according to claim 4, characterized in that, The specific method for monitoring the overall deformation of the air-base layer is as follows: A UAV equipped with a lidar acquires high-density three-dimensional point cloud data; a high-precision three-dimensional geometric model of the spatial structure is generated through point cloud filtering, registration, and modeling algorithms; based on point cloud differential analysis, the deformation at any given time is calculated. t Compared with reference state t The geometric offset of 0 enables the identification of changes in the overall arch of the roof and the displacement of key components. At the same time, a dynamic database of the evolution of the structural morphology over time is established through multi-temporal point cloud registration.

6. The method for monitoring the health of space structures based on sky-ground coordination according to claim 4, characterized in that, The specific method for identifying the apparent defects in the hollow base layer is as follows: The YOLO series target detection network is used to analyze high-definition images collected by UAVs. During a forward propagation, the category labels of apparent diseases and their rectangular bounding boxes are output to achieve the initial screening of diseases in large-area images. The candidate regions output by YOLO are used as regions of interest and input into a pixel-level segmentation network to perform fine segmentation of the disease boundary and generate a pixel-level mask. The pixel-level mask is used to quantify various surface defects.

7. The method for monitoring the health of space structures based on sky-ground coordination according to claim 1, characterized in that, At the grassroots level, a multi-dimensional sensing system is formed, supported by DIC displacement monitoring, MEMS multi-parameter environmental and dynamic sensing, and independent mechanical monitoring, and unified access is achieved through wireless network.

8. The method for monitoring the health of space structures based on sky-ground coordination according to claim 1, characterized in that, The steps for constructing the comprehensive risk index assessment model and implementing a three-tiered early warning system are as follows: Environmental threat level E Overall health H With local security S The indicators from the three dimensions are uniformly transformed into standardized features to construct the comprehensive risk index assessment model: ； in, α , β , γ The weighting coefficients are determined by expert weighting or the analytic hierarchy process, satisfying 𝛼+𝛽+𝛾=1; Based on meteorological and remote sensing image data acquired from surface-level monitoring, external load parameters are extracted, and an environmental risk index is calculated using a normalization method to obtain the environmental threat level. E The overall health score is obtained by forming a global health index based on point cloud data acquired from the spatial layer monitoring through fuzzy comprehensive evaluation. H Based on the number of times local monitoring parameters exceeded limits, cable imbalance coefficient, and cumulative fatigue damage obtained from ground level monitoring, the local safety margin of key components is evaluated to obtain the local safety margin. S ; Comprehensive risk index R The results are compared with the tiered thresholds to generate multi-level early warning results.

9. A space structure health monitoring system based on sky-ground coordination, characterized in that, It includes a multi-dimensional monitoring module, a data processing module, a performance evaluation module, a risk warning module, and an intelligent decision-making module; among which, The multi-dimensional monitoring module is used to monitor the spatial structure in the three dimensions of sky, air, and ground respectively, and obtain multi-dimensional monitoring information. The data processing module is used to perform spatiotemporal synchronization of the multi-dimensional monitoring information to obtain panoramic data; The performance evaluation module is used to comprehensively evaluate the static and dynamic performance of spatial structures and predict their life evolution trends based on the ground-level sensing data structure. The risk warning module is used to extract and fuse multi-source data features from the panoramic data, construct a comprehensive risk index assessment model, and conduct three-level graded warnings. The intelligent decision-making module is used to introduce an LLM model after an early warning is triggered and automatically generate a response plan based on the early warning structure.

10. A space structure health monitoring system based on sky-ground coordination according to claim 9, characterized in that, The multi-dimensional monitoring module, the data processing module, the performance evaluation module, the risk warning module, and the intelligent decision-making module are integrated through a lightweight integrated platform and connected to a multi-terminal collaborative management module.

Images