Crane multi-dimensional data dynamic analysis and visualization method, system and terminal device

Abstract

The application relates to the technical field of artificial intelligence, and provides a tower crane multi-dimensional data dynamic analysis and visualization method, system and terminal device. The method comprises the following steps: acquiring multi-dimensional monitoring data of a tower crane; a synchronous monitoring model of the tower crane is established based on the multi-dimensional monitoring data by using a digital twin technology, and the synchronous monitoring model is synchronously displayed in a user interface; based on a viewing instruction of the synchronous monitoring model by a user, a target display dimension is matched for the user, a folding dimension that does not match the viewing instruction is synchronously set, and model display information corresponding to the folding dimension is hidden in the synchronous monitoring model; a preset analysis model matched with the target display dimension is called, target monitoring information is extracted from the synchronous monitoring model, the target monitoring information is subjected to tower crane dynamic analysis by the preset analysis model; and the tower crane dynamic analysis result is updated to the synchronous monitoring model, so that model display information matched with the target display dimension in the synchronous monitoring model is updated, and the efficiency and accuracy of operation and maintenance decision-making are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a method, system, and terminal equipment for dynamic analysis and visualization of multidimensional data of tower cranes. Background Technology

[0002] In modern construction engineering, port logistics and heavy industry, tower cranes are core heavy lifting equipment, and their operating efficiency and safety directly determine the overall project progress, cost control and the safety of personnel and property.

[0003] With the continuous expansion of engineering projects and the increasing complexity of operating environments, tower crane operating conditions exhibit significant dynamism and multi-dimensional correlations. These involve the interaction and influence of various data aspects, including the equipment's mechanical performance, real-time operating loads, operator behavior, surrounding environmental conditions (such as wind speed, visibility, and obstacle distribution), equipment periodic inspection and maintenance records, and the execution of various operating commands. However, current tower crane management and monitoring systems generally suffer from fragmented data collection, outdated analysis methods, and limited visualization, making it difficult to meet the needs of dynamic management and control throughout the entire lifecycle of tower cranes under complex operating conditions. Furthermore, current tower crane monitoring data is mostly displayed using manually created data tables or charts, failing to organically integrate and intuitively present various data sources (such as equipment operating data, personnel operation data, and environmental data). This results in operators struggling to quickly and accurately grasp the overall operating status of the tower crane, often leading to delayed responses to tower crane status changes.

[0004] In summary, there is an urgent need to design a technical solution to address at least one of the following technical problems in related technologies: fragmented data acquisition, outdated analysis methods, limited visualization, and delayed response. Summary of the Invention

[0005] In this context, the embodiments of this application aim to provide a method, system, and terminal device for dynamic analysis and visualization of multidimensional data of tower cranes, thereby solving at least one technical problem existing in the related art.

[0006] In a first aspect of the embodiments of this application, a method for dynamic analysis and visualization of multidimensional data of tower cranes is provided, the method comprising:

[0007] Acquire multi-dimensional monitoring data of the tower crane; the multi-dimensional monitoring data includes at least: tower crane structural data, tower crane movement trajectory, environmental monitoring data, and tower crane image data;

[0008] Based on the multidimensional monitoring data, a synchronous monitoring model of the tower crane is established using digital twin technology, and the synchronous monitoring model is displayed synchronously in the user interface.

[0009] Receive user's viewing instruction for the synchronous monitoring model, match the target display dimension for the user based on the viewing instruction, synchronously set the folded dimension that does not match the viewing instruction, and hide the model display information corresponding to the folded dimension in the synchronous monitoring model;

[0010] A preset analysis model matching the target display dimension is invoked, target monitoring information is extracted from the synchronous monitoring model, and tower crane dynamic analysis is performed on the target monitoring information through the preset analysis model to obtain tower crane dynamic analysis results; wherein the preset analysis model includes one or a combination of the following models: anomaly detection model, fatigue life prediction model, and tower crane operation risk early warning model;

[0011] The tower crane dynamic analysis results are updated to the synchronous monitoring model to update the model display information that matches the target display dimension in the synchronous monitoring model.

[0012] In a second aspect of the embodiments of this application, a multi-dimensional data dynamic analysis and visualization system for tower cranes is provided, the system comprising the following modules:

[0013] The acquisition module is used to acquire multi-dimensional monitoring data of the tower crane; the multi-dimensional monitoring data includes at least: tower crane structural data, tower crane movement trajectory, environmental monitoring data, and tower crane image data;

[0014] The modeling module is used to establish a synchronous monitoring model of the tower crane based on the multidimensional monitoring data using digital twin technology, and then transmit it to the user interface;

[0015] A user interface is used to display the synchronous monitoring model; the model display information corresponding to the folded dimensions is hidden in the synchronous monitoring model;

[0016] The selection module is used to receive the user's viewing instruction for the synchronous monitoring model, match the target display dimension for the user based on the viewing instruction, and synchronously set the folded dimension that does not match the viewing instruction;

[0017] The analysis module is used to call a preset analysis model that matches the target display dimension, extract target monitoring information from the synchronous monitoring model, and perform tower crane dynamic analysis on the target monitoring information through the preset analysis model to obtain tower crane dynamic analysis results; wherein the preset analysis model includes one or a combination of the following models: anomaly detection model, fatigue life prediction model, and tower crane operation risk early warning model; and update the tower crane dynamic analysis results to the synchronous monitoring model to update the model display information in the synchronous monitoring model that matches the target display dimension.

[0018] This application discloses a method, system, and terminal device for multi-dimensional dynamic analysis and visualization of tower crane data. The embodiments of this application achieve full-dimensional, high-precision perception of the tower crane's operating status and customized display based on viewing commands. It solves the technical problems of fragmented data acquisition, outdated analysis methods, limited visualization, and delayed response in related technologies. Furthermore, through dynamic adaptation and intelligent analysis, it improves the efficiency and accuracy of operation and maintenance decisions, reduces safety risks during tower crane operation, and lowers manual operation and maintenance costs. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for dynamic analysis and visualization of multidimensional data of tower cranes as shown in this application;

[0020] Figure 2 This is a schematic diagram of the structure of a tower crane multidimensional data dynamic analysis and visualization system shown in this application. Detailed Implementation

[0021] The following is for reference. Figure 1 , Figure 1 This is a flowchart illustrating a method for dynamic analysis and visualization of multidimensional data of tower cranes, provided in one embodiment of this application.

[0022] To address at least one of the aforementioned technical problems, embodiments of this application provide a method, system, and terminal device for dynamic analysis and visualization of multi-dimensional tower crane data. Specifically, the embodiments of this application employ a multi-dimensional data collaboration, dynamic dimension adaptation, and real-time intelligent insight capability intelligent monitoring and visualization mode for tower cranes, solving the lagging issues of traditional tower crane monitoring methods that rely on local data collection, fixed-dimensional display, and manual-led analysis. Specifically, by integrating tower crane structural data, tower crane movement trajectory, environmental monitoring data, and tower crane image data to construct a full-dimensional data source, the limitations of data fragmentation and one-sided information in traditional monitoring are avoided. Relying on digital twin technology to establish a mapping between the physical tower crane and a virtual model, a synchronous monitoring model is generated, and differentiated visualization forms are adapted according to data types, such as structural data heatmaps, dynamic annotation of movement trajectories, and target embedding in image data, transforming the tower crane's operating status from abstract data into an intuitive scene, significantly reducing the complexity of data interpretation in traditional monitoring. Furthermore, the embodiments of this application dynamically match the target display dimension and hide / fold dimensions by receiving user viewing commands, avoiding interference from irrelevant data on the monitoring focus and achieving a personalized monitoring experience with on-demand focus. Simultaneously, pre-defined analysis models such as anomaly detection, fatigue life prediction, and operational risk warning are invoked to deeply analyze the target monitoring information extracted from the synchronous monitoring model. This replaces the subjectivity and lag of traditional manual judgment, enabling real-time identification of structural anomalies, prediction of component remaining life, and warning of collision risks, thus constructing a closed-loop chain from data acquisition to intelligent decision-making. The digital twin model provides a high-fidelity virtual verification scenario for dynamic analysis, capable of rehearsing the correlation between analysis results and the actual state of the tower crane under different monitoring dimensions, effectively avoiding safety hazards and maintenance losses caused by directly relying on trial and error with the physical tower crane. The dynamic analysis results are updated to the synchronous monitoring model in real time, ensuring consistency between the virtual display and the physical tower crane's operating status. At the same time, key conclusions are presented intuitively through visualization (such as anomaly location markings and risk level prompts), balancing the comprehensiveness of monitoring and the efficiency of decision-making, avoiding information overload or omission of key information. Ultimately, the implementation method of this application achieves full-dimensional and high-precision perception of tower crane operating status and customized display based on viewing instructions. It solves the technical problems of fragmented data collection, lagging analysis methods, single visualization presentation and delayed response in related technologies. Furthermore, it improves the efficiency and accuracy of operation and maintenance decision-making through dynamic adaptation and intelligent analysis, reduces safety risks in tower crane operation, and reduces manual operation and maintenance costs.

[0023] In practical applications, this embodiment of the application can also leverage artificial intelligence middleware to achieve core support for multi-dimensional data integration and efficient data flow. As a hub connecting heterogeneous data from multiple sources, such as tower crane structural data, motion trajectory data, environmental monitoring data, and image data, with upper-level digital twin models and intelligent analysis modules, it effectively solves the problem of data fragmentation in traditional monitoring. With multi-modal data preprocessing, dynamic adaptation, and intelligent scheduling capabilities, the middleware achieves efficient integration and real-time synchronization of various dispersed data, providing a stable data transmission channel for the accurate mapping of physical tower cranes and virtual models, and the construction of synchronous monitoring models. It also supports dynamic dimensional adaptation and rapid response to viewing commands, meeting the closed-loop requirements of this embodiment from data acquisition to intelligent decision-making.

[0024] Furthermore, this application embodiment also transforms technologies such as digital twin technology, dynamic adaptation logic, and preset analysis models into an actual system that can be run on terminal devices through software development. Ultimately, it realizes core functions such as customized display based on viewing instructions and intuitive presentation of key conclusions (abnormal location marking, risk level prompts), directly achieving the goal of reducing manual operation and maintenance costs and improving decision-making efficiency and accuracy. It is a concrete manifestation of the application of intelligent management of tower cranes.

[0025] Figure 1 The flowchart of a method for dynamic analysis and visualization of multidimensional data of tower cranes, as shown in an embodiment of this application, includes:

[0026] Step S101: Obtain multi-dimensional monitoring data of the tower crane;

[0027] Step S102: Based on the multidimensional monitoring data, a synchronous monitoring model of the tower crane is established using digital twin technology, and the synchronous monitoring model is displayed synchronously in the user interface.

[0028] Step S103: Receive the user's viewing instruction for the synchronous monitoring model, match the target display dimension for the user based on the viewing instruction, synchronously set the folded dimension that does not match the viewing instruction, and hide the model display information corresponding to the folded dimension in the synchronous monitoring model;

[0029] Step S104: Call the preset analysis model that matches the target display dimension, extract the target monitoring information from the synchronous monitoring model, and perform tower crane dynamic analysis on the target monitoring information through the preset analysis model to obtain the tower crane dynamic analysis result;

[0030] Step S105: Update the tower crane dynamic analysis results to the synchronous monitoring model to update the model display information matching the target display dimension in the synchronous monitoring model.

[0031] In this embodiment of the application, the multidimensional monitoring data includes at least: tower crane structural data, tower crane movement trajectory, environmental monitoring data, and tower crane image data.

[0032] Specifically, tower crane structural data comprises mechanical and physical parameters reflecting the stress state and health of the main structure of the tower crane. These parameters are directly related to the structural safety of the tower crane and are crucial for assessing structural fatigue and damage risks. This type of data primarily monitors the mechanical characteristics and physical state of the load-bearing components of the tower crane, including real-time stress values, minute deformations, vibration characteristics, and temperatures of critical components. Stress values ​​encompass tensile and compressive stresses, measured in megapascals (MPA). It is necessary to distinguish between static stress generated by the structure's own weight and dynamic stress generated during the lifting and rotation of the load. Minor deformations, i.e., strain, measured in microstrain, can indirectly deduce stress distribution, avoiding localized biases from single-stress monitoring. Vibration characteristics include vibration acceleration and vibration frequency, measured in meters per second squared (m²) and Hertz (Hz), respectively. Special attention should be paid to the risk of boom swaying vibration and tower resonance; peak vibration values, duration, and triggering scenarios must be recorded. Temperatures of critical components are primarily monitored at high-strength bolt connection points and weld areas, measured in degrees Celsius, to prevent extreme temperatures from causing changes in material mechanical properties that could affect structural safety.

[0033] Tower crane structural data is collected by deploying dedicated sensors at key locations. Stress and strain data rely on resistance strain gauges or fiber optic grating sensors; the former is low-cost and easy to deploy, while the latter has strong electromagnetic interference resistance, high accuracy, and is suitable for long-term monitoring. These sensors are either attached to the surface of components or embedded internally. Vibration data is collected by triaxial accelerometers, which are fixed at vibration-sensitive points such as the middle of the boom and the top of the tower. Temperature data is acquired through patch-type temperature sensors, which are deployed in the same location as stress sensors, enabling stress-temperature correlation analysis. By monitoring tower crane structural data, abnormal structural stresses, such as sudden increases in stress and excessive strain, can be detected in real time, providing early warnings of potential faults such as boom cracking and tower deformation. Combining historical data can also analyze the cumulative fatigue patterns of the structure, providing precise basis for component maintenance, such as determining the timing of bolt tightening and weld inspection, ultimately avoiding the risk of structural collapse.

[0034] Tower crane motion trajectory data describes the position, speed, and trend of each actuator of the tower crane. It is used to determine the tower crane's operating posture and is core data for collision avoidance and operational efficiency assessment. This type of data covers the motion characteristics of the tower crane's slewing, luffing, and hoisting mechanisms, as well as the coordinated state when multiple mechanisms are linked. Specifically, tower crane motion trajectory data is collected through position and speed sensors deployed on each motion mechanism. The position data for the slewing, luffing, and hoisting mechanisms relies on absolute encoders, which are directly connected to the mechanism's drive shaft and have no cumulative error. In some scenarios, GPS or BeiDou positioning modules are also added for overall tower crane position calibration, such as relative position positioning during multi-tower operations. Speed ​​and acceleration data are acquired through incremental encoders or Hall effect sensors. Incremental encoders derive speed by calculating the number of pulses per unit time, while Hall effect sensors indirectly convert speed and acceleration by detecting motor speed. Using tower crane motion trajectory data, the tower crane's operating trajectory can be accurately recorded, providing positional basis for collision avoidance in multi-tower operations, such as determining whether the hooks of adjacent tower cranes have entered a safe distance. By analyzing motion speed and acceleration, operating habits can be optimized, such as avoiding sudden acceleration that could cause the load to sway. At the same time, it can also count the operating time of each mechanism, such as the total rotation angle and the total lifting height, and then evaluate the equipment operating load and operating efficiency.

[0035] Tower crane environmental monitoring data are parameters reflecting the surrounding natural environment and on-site working conditions. They are crucial for determining whether the working environment meets safety requirements and deciding whether to cease operations. This data covers both natural environmental conditions and on-site operational interference factors, directly impacting tower crane operation safety and stable equipment operation.

[0036] Tower crane environmental monitoring data is collected through environmental sensors deployed on the top of the tower crane, in the operator's cab, and at the work site. Wind data relies on cup anemometers or ultrasonic anemometers. Cup anemometers are low-cost, while ultrasonic anemometers have no mechanical wear and are suitable for long-term outdoor use. These instruments are installed on the unobstructed top of the tower crane. Temperature, humidity, and rainfall data are obtained through integrated temperature, humidity, and rainfall sensors, which are installed in a rainproof location on the top of the operator's cab. Visibility and distance data to surrounding obstacles are collected through laser rangefinders or visibility meters. Laser rangefinders are used to monitor the distance to fixed obstacles, while visibility meters are used in foggy conditions. In some scenarios, tower crane image data is also used to assist in judging environmental conditions. Monitoring tower crane environmental data can identify severe environmental risks in real time, such as strong winds, heavy rain, and low visibility, thereby triggering automatic warnings, such as triggering audible and visual alarms and prohibiting lifting operations when wind speeds exceed limits. Recording environmental data can also trace the causes of faults, such as electrical faults caused by high humidity. It can also provide a basis for work planning, such as scheduling lifting operations to avoid strong wind periods.

[0037] Tower crane image data is visual data of tower crane operation scenes collected by cameras, covering visual information such as the status of the hoisted object, the surrounding environment, and the appearance of structural components. It is used to supplement blind spots in sensor data and achieve visual monitoring. This type of data covers key tower crane operation scenarios and can intuitively present details that sensors cannot capture. Specifically, image data related to hoisted object and hook monitoring includes real-time video streams and periodic snapshots. The real-time video stream resolution is no less than 1080P, and the frame rate is no less than 25 frames per second. Periodic snapshots are taken at one frame every 3 to 5 seconds, clearly showing the hoisted object's binding status, type, and the effectiveness of the hook's anti-derailment device, such as whether the wire rope used for binding the hoisted object is loose, and whether the hoisted object is a steel or concrete component. Image data related to boom and slewing radius monitoring is a real-time video stream, capturing the scene below and at the end of the boom. It can monitor whether the boom is in contact with surrounding obstacles and observe whether personnel have accidentally entered within the slewing radius, and must support moving target recognition functionality. The monitoring data related to the tower body and foundation consists of periodic image captures, typically once per hour, capturing images of the bottom of the tower body and the foundation platform. This helps determine the tower's verticality and whether there are cracks or water accumulation in the foundation. The monitoring data related to the operator's cab blind spot is a real-time video stream. For areas not covered by the operator's cab's view, such as the rear of the tower crane or behind the luffing trolley, side-mounted cameras capture images to eliminate blind spots.

[0038] For example, tower crane image data can intuitively present visual information that sensors cannot capture, such as loose loads or personnel accidentally entering dangerous areas, triggering timely manual intervention. Combined with image recognition algorithms, such as YOLO and Faster R-CNN, anomalies can be automatically identified, such as tilted loads or obstructions, improving the intelligence level of monitoring. Simultaneously, image data can also serve as evidence for accident tracing, such as reconstructing abnormal scenarios during hoisting operations.

[0039] As an optional embodiment, in step S101, acquiring multi-dimensional monitoring data of the tower crane can be achieved by: collecting tower crane operation data in multiple dimensions through various types of sensors and acquisition devices deployed at key parts of the tower crane, and completing preliminary data aggregation and format unification through an edge computing gateway. Specifically, high-precision stress and strain sensors (such as resistance strain gauges and fiber optic grating sensors) are installed at key stress-bearing parts such as the standard tower section, the root and end of the boom, and the hook beam, and the acquisition frequency is set to collect structural stress and strain data in real time. At the same time, vibration acceleration data of key components is collected through vibration sensors and transmitted to the edge gateway for backup. Absolute encoders are installed on the tower crane's slewing mechanism, luffing trolley drive mechanism, and hoisting mechanism to collect slewing angle, trolley displacement, and hook lifting height data, and the acquisition frequency is set. Simultaneously, a GPS / BeiDou positioning module is installed on the top of the tower crane to assist in calibrating the slewing center position and the real-time spatial coordinates of the boom, forming a complete motion trajectory dataset. Wind speed and direction sensors, as well as temperature and humidity sensors, are installed at the top railing of the tower crane. A rainfall sensor is installed near the operator's cab (for outdoor operations). Real-time data on wind speed, wind direction, ambient temperature, humidity, and rainfall are acquired at a frequency of 1-5Hz. The sensors are waterproof and dustproof to ensure stable data acquisition in outdoor environments. High-definition network cameras are installed below the end of the tower crane boom, above the hook, around the base of the tower, and in the blind spots of the operator's cab. These cameras support low-light shooting to adapt to nighttime operations. Real-time video streams are transmitted to the edge gateway via the RTSP protocol, while simultaneously triggering the cameras to periodically capture static images for subsequent target detection preprocessing. After receiving the raw data from each sensor and device, the edge computing gateway encapsulates it according to a preset data format. Abnormal missing data from offline sensors is marked as invalid, and abnormal data exceeding the physical range (such as negative stress values ​​or wind speeds exceeding 50m / s) undergoes preliminary filtering. Raw data frames containing data acquisition timestamps, sensor numbers, and data types are generated as the basic input for spatiotemporal alignment and noise reduction processing.

[0040] As an optional embodiment, in step S102, a synchronous monitoring model of the tower crane is established using digital twin technology based on the multidimensional monitoring data, and the synchronous monitoring model is displayed synchronously in the user interface. This includes: performing spatiotemporal alignment and denoising on the tower crane structural data, tower crane motion trajectory, environmental monitoring data, and tower crane image data in the multidimensional monitoring data using Kalman filtering and particle filtering algorithms to generate a fused data frame; constructing a digital twin model containing a geometric model, a physical model, and an environmental model based on the fused data frame; establishing a structural health status mapping model using a graph convolutional network based on the geometric model and the physical model; calculating the reconstruction error through the structural health status mapping model to detect whether the tower crane is in good condition. If structural anomalies are found, the output includes a tower crane structural safety assessment result containing reconstruction errors and structural anomaly detection results. Based on a digital twin model, keyframe interpolation is performed on the dynamic attribute data in the fused data frame in the GPU shader. Based on the tower crane structural safety assessment result, the corresponding abnormal change area in the digital twin model is located. The interpolation calculation result is updated through an event-driven incremental rendering algorithm to obtain the real-time rendering load. The real-time rendering load is displayed in the user interface. Combining different data types in the fused data frame, the non-dynamic attribute data in the fused data frame is adapted to a visualization format and synchronously rendered to the user interface according to the adapted visualization format to obtain the synchronous monitoring model.

[0041] Specifically, in step S102, the process of establishing a tower crane synchronous monitoring model using digital twin technology based on multi-dimensional monitoring data and synchronously displaying it on the user interface needs to be carried out in an orderly manner through multiple stages. The core is to achieve synchronization between the physical tower crane and the virtual model through the collaboration of data preprocessing, model building, anomaly detection, and visualization rendering. First, in step S102, multi-dimensional monitoring data preprocessing is performed. For the tower crane structural data, tower crane movement trajectory, environmental monitoring data, and tower crane image data included in the multi-dimensional monitoring data, Kalman filtering and particle filtering algorithms are used together for processing. The two algorithms work together to achieve spatiotemporal alignment of data from different sources, ensuring that various types of data maintain consistency in the time and space dimensions, and eliminating spatiotemporal deviations caused by sensor response delays and differences in installation positions during data acquisition. On the other hand, they filter out noise interference in the data, such as random errors of the sensors themselves and abnormal data points caused by external electromagnetic interference, ultimately generating a fused data frame that integrates various effective information, providing a high-quality data foundation for subsequent model building.

[0042] Next, in step S102, a digital twin model is constructed based on the fused data frame. This model comprises three core components: a geometric model, a physical model, and an environmental model. The geometric model primarily recreates the physical structure of the tower crane, constructed based on information such as the tower crane's structural dimensions and component positions from the fused data frame. The physical model focuses on simulating the tower crane's mechanical behavior and motion characteristics, combining structural stress, strain data, and motion trajectory parameters from the fused data frame to reproduce the tower crane's stress state and mechanical actions during operation. The environmental model integrates real-time environmental data from the fused data frame, such as wind speed, temperature, and rainfall, to construct a virtual environment consistent with the actual operating scenario of the tower crane. The three components are then integrated to form a complete digital twin model.

[0043] Subsequently, in step S102, a structural health status mapping model is established based on the geometric model and the physical model. This process is implemented using a graph convolutional network. First, the physical connection relationships of tower crane components in the geometric model and the virtual sensor monitoring points in the physical model are abstracted into a graph structure, where the connection nodes of tower crane components and the sensor monitoring points are the nodes of the graph, and the physical connections and data transmission relationships between components are the edges of the graph. Then, the graph convolutional network learns the characteristics of each node in the graph structure and the correlation rules between nodes, thereby constructing a mapping model that reflects the structural health status of the tower crane. Afterward, the reconstruction error is calculated through this mapping model, that is, the difference between the structural status data output by the model and the actual structural data in the fused data frame is compared. Based on the magnitude of the reconstruction error, it is determined whether there is a structural anomaly in the tower crane. For example, if the error exceeds a preset range, it is determined that there is a risk of structural damage. Finally, the tower crane structural safety assessment result containing the specific reconstruction error value and the structural anomaly detection result is output.

[0044] In the rendering and display phase, based on the constructed digital twin model, the dynamic attribute data in the fused data frames is first processed. This dynamic attribute data mainly includes parameters that change dynamically over time, such as the slewing angle and trolley displacement in the tower crane's trajectory, and real-time stress changes in the structural data. Keyframe interpolation calculations are performed on this data using GPU shaders to fill in parameter gaps between adjacent data acquisition moments, making the dynamic changes of the virtual model smoother and more coherent. Simultaneously, combined with the previously output tower crane structural safety assessment results, the regions in the digital twin model corresponding to structural anomalies are located. Then, an event-driven incremental rendering algorithm is used to update the keyframe interpolation calculation results. Only the abnormal change areas and the parts of the dynamic attribute data that have changed are rendered and updated, rather than rendering the entire model, reducing the rendering load and obtaining a real-time rendering load that reflects the tower crane's status in real time. This real-time rendering load is then displayed on the user interface.

[0045] Understandably, in updating the keyframe interpolation results based on the tower crane structural safety assessment results, the precise location of structural anomaly areas must first be achieved. The tower crane structural safety assessment results include information such as the specific location, type, and severity of structural anomalies. This information has a clear correspondence with the geometric and physical models in the digital twin model. Specifically, the anomaly location mentioned in the assessment results can be mapped to a specific component in the geometric model (such as a standard tower section or a specific section of the boom), while the anomaly type is associated with the force or deformation data of virtual sensor monitoring points in the physical model. Through this mapping, a spatial region that perfectly matches the structural anomaly is located in the digital twin model, clarifying the 3D coordinate range of this region in the virtual model and the corresponding component identifier, thus defining precise boundaries for subsequent rendering updates. Subsequently, based on the located anomaly region, an event-driven incremental rendering algorithm is initiated to update the keyframe interpolation results. The detection of the structural anomaly itself constitutes an event that triggers the rendering update; the algorithm only responds to the anomaly region corresponding to this event, rather than processing the entire digital twin model. First, historical data of the abnormal region in keyframe interpolation calculations is extracted. Then, combined with the latest abnormal data from the structural safety assessment results (such as real-time stress changes and deformation increments), the keyframe interpolation parameters for this region are recalculated to generate new interpolation results that only cover the abnormal region, ensuring that the interpolation data reflects the dynamic changes of structural anomalies. For example, in the change region location stage, a difference analysis is performed on the keyframe interpolation calculation results corresponding to the triggering event. First, two versions of keyframe interpolation data before and after the event are extracted. By comparing the data, the parameter dimensions whose values ​​have changed are identified, such as the change in the trolley displacement interpolation results within a certain time period and the abnormal fluctuation of the boom stress interpolation values. Then, combined with the structural topology relationship of the digital twin model, these changed parameter dimensions are mapped to the spatial region of the virtual model to accurately locate the local areas that need to be rendered and updated, such as the new movement position of the luffing trolley and the specific section on the boom where the stress exceeds the standard. At the same time, areas with no change in values ​​are excluded to ensure that subsequent updates only focus on the necessary parts. Next, the newly generated local interpolation results are fused with the original interpolation results of the non-abnormal areas in the model. The visual discontinuity between regions is eliminated through edge transition processing. Finally, only the rendering content of the abnormal areas in the user interface is updated, which ensures the real-time presentation of the abnormal state, greatly reduces the consumption of rendering resources, and maintains the smoothness of the model display.

[0046] Finally, in step S102, the non-dynamic attribute data in the fused data frame is processed. This type of data includes fixed structural dimensions of the tower crane, component models, sensor installation locations, and other data that does not change or changes slowly over time. Based on the different types of this non-dynamic attribute data, corresponding visualization formats are adapted. For example, structural dimension data is presented using 3D model annotations, and component model data is displayed using text pop-ups. Then, according to the adapted visualization format, the non-dynamic attribute data is synchronously rendered to the user interface and integrated with the previously displayed real-time rendering load to ultimately form a complete synchronous monitoring model, achieving real-time correspondence between the tower crane's physical state and the virtual model display.

[0047] Further optionally, in one embodiment of step S102, constructing a digital twin model comprising a geometric model, a physical model, and an environmental model based on the fused data frame includes: calling the tower crane structural dimensions and component position data from the fused data frame to construct a geometric model using a 3D modeling tool; using finite element analysis combined with the original structural stress and strain data from the fused data frame to simulate the structural mechanical behavior of the tower and boom in the tower crane, and using multibody dynamics combined with the motion trajectory parameters from the fused data frame to simulate the real-time motion trajectory of each moving component in the tower crane, fusing the structural mechanical behavior and real-time motion trajectory to obtain a physical model; integrating the real-time environmental data from the fused data frame to obtain an environmental model; and fusing the geometric model, physical model, and environmental model to form a digital twin model.

[0048] Specifically, when constructing a digital twin model that includes a geometric model, a physical model, and an environmental model based on a fused data frame, it is necessary to proceed in stages according to the model type. The construction of each model is based on the corresponding data in the fused data frame to ensure that the virtual model is highly consistent with the physical tower crane.

[0049] First, a geometric model is constructed, the core of which is to recreate the physical structural form of the tower crane. This step requires retrieving the tower crane's structural dimensions and component location data from the fused data frame. This data includes key structural parameters such as the standard section height of the tower crane, the boom length, and the hook beam dimensions, as well as the relative installation positions of each component. This data is then input into a 3D modeling tool, and through the tool's structural modeling function, the overall appearance and internal component layout of the tower crane are reproduced at a 1:1 scale, ultimately forming a geometric model. This model can intuitively present the physical structural characteristics of the tower crane, providing a basic framework for the subsequent integration of the physical model and the environmental model. Next, a physical model is constructed, focusing on simulating the mechanical behavior and motion characteristics of the tower crane, which needs to be completed in two parts. The first part simulates the structural mechanical behavior, using finite element analysis technology. The original structural stress and strain data from the fused data frame are used as input. This data reflects the stress conditions of the tower crane and boom under different operating conditions. Through finite element analysis, the tower crane and boom are mechanically modeled, calculating the stress distribution and deformation trends of components under different stress conditions, accurately reproducing their structural mechanical behavior. The second part simulates the real-time motion trajectory using multibody dynamics technology. It utilizes motion trajectory parameters from the fused data frames, including slewing angle, trolley displacement, and hook lifting height. Motion models of each moving component of the tower crane (slewing mechanism, luffing mechanism, and hoisting mechanism) are established through multibody dynamics, simulating the movement trajectory and speed changes of each component during operation. Finally, the structural mechanics simulation results are fused with the real-time motion trajectory simulation results to form a complete physical model that dynamically reflects the force and motion state of the tower crane during operation. Next, an environmental model is constructed, crucially recreating the actual operating environment of the tower crane. Real-time environmental data is directly extracted from the fused data frames, covering natural environmental parameters such as wind speed, wind direction, ambient temperature, humidity, and rainfall, as well as on-site working condition parameters such as the distance to surrounding obstacles. This data is integrated into a unified environmental modeling module, and through data mapping and scene construction, an environmental model consistent with the actual operating scenario of the tower crane is formed. This model provides the environmental background for the subsequent overall operation of the digital twin model, ensuring that the virtual simulation matches the actual operating environment conditions. Finally, the digital twin model is integrated, synergistically fusing the geometric model, physical model, and environmental model. Based on a geometric model, the mechanical and motion simulation functions of a physical model are embedded within it, giving the geometric model dynamic response capabilities. Simultaneously, the scene parameters of the environmental model are linked to the physical model, allowing the physical model to be affected by environmental factors during simulation, such as the interference of wind speed changes on the boom's movement trajectory. The integration of these three elements forms a digital twin model that comprehensively reflects the tower crane's structural morphology, mechanical behavior, motion state, and operating environment, providing a core virtual platform for subsequent synchronous monitoring and analysis.

[0050] Further optionally, in one embodiment of step S102, the virtual sensor monitoring points constructed in the physical model and the physical connection relationship constructed in the geometric model are converted into corresponding graph structures, and a graph convolutional network (GCN) is used to learn the topological features of each graph node in the graph structure and the topological features between graph nodes to establish a structural health status mapping model.

[0051] The above steps require first converting information from the physical and geometric models into a graph structure, then learning topological features through a graph convolutional network, and finally establishing a structural health status mapping model. The core is to transform the physical relationships of the tower crane structure into computable graph data, thereby uncovering structural health patterns. First, the graph structure is constructed, using virtual sensor monitoring points in the physical model and physical connections in the geometric model as core elements. Virtual sensor monitoring points are key locations in the physical model that simulate data collection by actual sensors. Each monitoring point corresponds to a specific stress or status monitoring location on the tower crane structure. These monitoring points serve as nodes in the graph structure, each carrying structural data features it has collected, such as stress and strain parameters. The physical connections in the geometric model correspond to edges in the graph structure. These connections represent the actual structural relationships between various components of the tower crane, such as bolted connections between standard tower sections and hinged connections between the boom and the tower cap. The existence of edges indicates that the corresponding node (virtual sensor monitoring point) has a physical force transmission or positional relationship within the structural component, thus connecting the scattered nodes into a complete graph structure according to the actual structural logic. Next, a graph convolutional network is used to learn features from the graph structure. Graph convolutional networks (GCNNs) can leverage the topological information of graph structures to overcome the limitations of traditional neural networks in processing grid data. During the learning process, the GCNN first reads the structural data features carried by each node, and then aggregates the feature information of neighboring nodes through specific convolution operations. That is, based on the edge connections, the features of the target node and its neighbors are fused and calculated. This fusion not only preserves the local state features of individual nodes but also captures the topological relationships between nodes. For example, how stress changes at a monitoring point of a tower standard section affect the state of adjacent standard sections through connections. Through multiple layers of such convolution operations, the GCNN can gradually learn the global topological features of the entire tower crane structure. These features encompass the association patterns and data change patterns of various parts of the structure. Finally, a structural health status mapping model is established based on the learned topological features. The topological features output by the GCNN are correlated with the known structural health status of the tower crane, including different levels such as normal, minor damage, moderate damage, and severe damage. The model learns the distribution patterns of topological features under different health states. For example, when the structure experiences minor damage, the features of the corresponding damaged nodes and the associated features of surrounding nodes will exhibit specific change patterns. After training, the structural health status mapping model can map the input graph structure topology features to the current structural health status of the tower crane. Subsequently, only the real-time constructed tower crane structure graph and corresponding features need to be input, and the model can output the corresponding structural health status assessment results.

[0052] Further optionally, in one embodiment of step S102, combining different data types in the fused data frame, the non-dynamic attribute data in the fused data frame is adapted to a visualization format, including: for point cloud data in the fused data frame, calling the PointLSTM model to predict motion trajectories, and embedding the motion trajectory prediction results into the synchronous monitoring model; for the tower crane structural safety assessment results output by the graph convolutional network, overlaying them into the corresponding structural components in the synchronous monitoring model in the form of a heatmap; for environmental monitoring data in the fused data frame, marking them in the synchronous monitoring model using a spatiotemporal grid method, and overlaying dynamic annotations; for image data in the fused data frame, performing target detection for obstacles and suspended objects using YOLO and Faster R-CNN, and embedding the target detection results into the corresponding viewpoint in the synchronous monitoring model.

[0053] Understandably, when adapting the visualization format for non-dynamic attribute data in the fused data frame, targeted processing methods must be adopted based on the characteristics and presentation requirements of different data types to ensure that various types of data are displayed intuitively and accurately in the synchronous monitoring model. For point cloud data in the fused data frame, the PointLSTM model is first called for processing. This model can learn the motion patterns of relevant tower crane components based on the spatial location and temporal change information contained in the point cloud data, and then predict their subsequent motion trajectories. After the prediction is completed, the obtained motion trajectory prediction results are embedded into the synchronous monitoring model in the form of dynamic lines, enabling the model to display the possible motion paths of the components in advance and providing a visual reference for operation prediction.

[0054] The tower crane structural safety assessment results output by the graph convolutional network are visualized using heatmaps. The color gradient of the heatmap corresponds to the structural safety status; for example, darker colors indicate higher structural safety risks, while lighter colors indicate more stable structures. These heatmaps are precisely overlaid onto the corresponding structural components in the synchronous monitoring model, allowing operators to intuitively identify weak areas or potential risk points in the tower crane structure and quickly grasp the overall structural safety status. Environmental monitoring data in the fused data frames is marked using a spatiotemporal grid. The spatiotemporal grid is divided into several units based on time and space dimensions, with each unit corresponding to environmental data for a specific time period and region. Different colors or symbols are used to mark the numerical ranges of environmental parameters within the grid; for example, different colors represent different wind speed ranges. Simultaneously, dynamic annotations are overlaid on top of the grid to display specific environmental parameter values ​​in real time, such as current temperature and humidity, making the spatiotemporal distribution characteristics of the environmental data clearly visible. For image data in the fused data frames, object detection is performed using both YOLO and Faster R-CNN algorithms. These algorithms can identify the location, outline, and type of obstacles and suspended objects from the images. The detected target information, such as the boundary range of obstacles and the real-time status of suspended objects, is embedded into the corresponding viewpoint in the synchronous monitoring model in the form of bounding boxes or labels. That is, a virtual viewpoint consistent with the image acquisition location, so that the virtual scene in the model corresponds to the target information in the actual image, eliminating visual blind spots and improving the comprehensiveness of monitoring.

[0055] Optionally, after step S102, to ensure that the virtual digital twin model remains consistent with the physical tower crane's state, dynamic correction and updating of the model are achieved through a closed-loop process of data feedback, algorithm calculation, and parameter adjustment. Specifically, the input data required for correction is first determined. On one hand, synchronous monitoring model status data fed back from the user interface is acquired. This data mainly reflects the deviation between the virtual model and the physical tower crane, such as the deviation between the motion trajectory simulated by the virtual model and the actual motion trajectory of the physical tower crane, and the deviation between the structural stress calculated by the virtual model and the actual structural stress of the physical tower crane. On the other hand, fused data frames collected and updated in real time by the edge computing gateway are acquired. This data directly reflects the latest state of the physical tower crane, providing real-time basis for deviation analysis and model adjustment.

[0056] Next, data processing and calculation are performed using two core algorithms. The first is a model predictive control algorithm, which takes the acquired deviation data and real-time fused data frames as input. Based on a preset model accuracy target, this algorithm calculates the correction amount that can reduce the virtual-physical deviation. These correction amounts correspond to the parameter ranges that need to be adjusted in the digital twin model, with the aim of making the output of the virtual model closer to the actual state of the physical tower crane. The second is a time-series predictive model combining Seq2Seq and Transformer. When this model is invoked, historically accumulated fused data and current real-time deviation data are used as input. Through the model's learning of time-series patterns, it predicts the future trends of the tower crane's motion, stress, and other states, providing a forward-looking reference for model adjustments and avoiding adjustment lag caused by relying solely on current deviations.

[0057] Then, the model parameters are fed back and adjusted. The correction values ​​output by the model predictive control algorithm and the future state parameters output by the time-series prediction model are used together as the model update parameters and synchronously fed back to the physical and environmental models of the digital twin model. For the physical model, its mechanical calculation parameters are adjusted according to the updated parameters, such as modifying the material coefficients in structural stress calculations and the resistance parameters in motion trajectory simulations, to ensure that the physical model more accurately simulates the mechanical behavior and motion state of the tower crane. For the environmental model, its environmental simulation parameters are adjusted according to the updated parameters, such as correcting the influence coefficient of wind speed on the boom stress and the simulation parameters of temperature on the structural material properties, so that the environmental model more closely reflects the actual operating environment's impact on the tower crane.

[0058] Finally, the model synchronization and display are completed. After parameter adjustment, the simulated state of the digital twin model is highly consistent with the actual state of the physical tower crane. At this time, the digital twin model is updated to the final synchronized monitoring model and pushed to the user interface as the latest model display information, ensuring that the virtual model seen by the user on the interface can reflect the operating status of the physical tower crane in real time and accurately.

[0059] As an optional embodiment, in step S103, receiving a user's viewing instruction for the synchronous monitoring model, matching a target display dimension for the user based on the viewing instruction, and synchronously setting a collapsed dimension that does not match the viewing instruction includes: receiving the user-initiated viewing instruction through a user interface, wherein the input form of the viewing instruction includes manual text input, voice command, and / or gesture operation. This viewing instruction contains the user's core needs for tower crane monitoring. Then, the viewing instruction is input to a Seq2Seq-based dynamic semantic parser, which parses the viewing instruction into a structured query to obtain key instruction information, establishing a mapping relationship between the key instruction information and each monitoring data category in the multi-dimensional monitoring data. Based on the mapping relationship, a graph attention network is used to evaluate the correlation between each monitoring data category and the key instruction information. Monitoring data categories with a correlation higher than a correlation threshold are selected as target display dimensions, and monitoring data categories with a correlation lower than a correlation threshold are marked as collapsed dimensions. A reinforcement learning agent is used to dynamically adjust the correlation threshold based on the context information and historical interaction feedback information of the viewing instruction. For example, when the command is to assess collision risk, the tower crane's movement trajectory and image data are identified as the target display dimensions, while the tower crane's structural data and environmental monitoring data are marked as collapsible dimensions. Next, combining historical interaction databases, the K-means clustering algorithm is used to perform cluster analysis on users' past viewing dimension preferences. Finally, based on the clustering analysis results, the target display dimensions and collapsible dimensions are optimized and adjusted, retaining monitoring data categories whose user attention frequency reaches a preset condition as the final output target display dimensions, and using monitoring data categories whose user attention frequency does not reach the preset condition as collapsible dimensions.

[0060] Specifically, in step S103, the process of receiving user viewing instructions and matching target display dimensions and setting collapsed dimensions requires the coordinated efforts of semantic parsing, relevance assessment, threshold adjustment, and preference analysis to achieve adaptive matching and personalized optimization of dimensions. First, the user-initiated viewing instructions are received. These instructions are input through the user interface, including manually entered text, voice-converted commands, and gestures. These instructions directly reflect the user's core needs for tower crane monitoring, such as monitoring structural safety status or collision risk during operations. Next, the viewing instructions are semantically parsed and a mapping relationship is established. The received instructions are input into a Seq2Seq-based dynamic semantic parser. This parser deeply understands the semantic logic of the instructions, transforming unstructured instructions into structured queries and extracting key information that directly corresponds to the user's monitoring focus. Subsequently, based on this key information, a mapping relationship is established between it and various monitoring data categories in the multi-dimensional monitoring data, identifying which data categories might be relevant to the user's needs. Then, the relevance between the data categories and the key information of the instructions is evaluated using a graph attention network based on the established mapping relationship. Graph attention networks can focus on data category features closely related to key information in instructions, calculating and outputting the relevance between each monitored data category and the key information. Based on the calculation results, monitored data categories with relevance higher than a preset relevance threshold are selected and identified as target display dimensions. Monitored data categories with relevance not higher than the threshold are marked as collapsed dimensions. Simultaneously, a reinforcement learning agent dynamically adjusts the relevance threshold. This agent learns the threshold adjustment pattern based on the context information of the viewing instruction (i.e., the background needs associated with the instruction) and historical interaction feedback information (i.e., the user's past adjustment records of dimension display), enabling the threshold to adapt to the relevance judgment needs under different instruction scenarios and improving the accuracy of dimension selection. Next, combining historical interaction database analysis of user preferences, K-means clustering algorithm is used to cluster and analyze the user's past viewing dimension preferences, classifying the dimensions selected multiple times by the user and identifying the user's typical attention dimension patterns in different scenarios. Finally, the target display dimensions and collapsed dimensions are optimized and adjusted based on the clustering analysis results. For monitored data categories whose user attention frequency reaches the preset condition, even if their relevance is slightly lower than the threshold, they are still retained as the final target display dimensions. Categories whose user attention frequency does not meet the preset conditions may be adjusted to collapse dimensions, even if they are slightly more relevant, to ultimately form a dimension display scheme that meets the user's personalized needs.

[0061] In this application embodiment, the preset analysis model includes one or a combination of the following models: anomaly detection model, fatigue life prediction model, and tower crane operation risk early warning model.

[0062] As an optional embodiment, in step S104, a preset analysis model matching the target display dimension is invoked, target monitoring information is extracted from the synchronous monitoring model, and tower crane dynamic analysis is performed on the target monitoring information through the preset analysis model to obtain the tower crane dynamic analysis result. This includes: when the data type corresponding to the target display dimension is tower crane structural data, tower crane structural data is extracted from the synchronous monitoring model as target monitoring information, the target monitoring information including stress, strain data, and structural health status mapping model data; an anomaly detection model and fatigue life prediction model are invoked as matching preset analysis models; and 1D-CNN is used to monitor the target through the anomaly detection model. Local features are extracted from the time-series structural data in the monitoring information, and the Canopy-KMeans two-level clustering algorithm is used to identify tower crane anomaly patterns in the extracted local features, resulting in structural anomaly detection results that include the abnormal structure, its location, and the type of anomaly. Through a fatigue life prediction model, the stress time-series data from the target monitoring information is input into an LSTM to learn the fatigue accumulation law of the tower crane under cyclic loads. The remaining fatigue life prediction value and corresponding fatigue accumulation factors are predicted by combining a BP neural network and a finite element sample analysis algorithm. Finally, the structural anomaly detection results, the remaining fatigue life prediction value, and the corresponding fatigue accumulation factors are integrated to form a dynamic analysis result for the tower crane structural data.

[0063] Specifically, when the data type corresponding to the target display dimension is tower crane structural data, in step S104, the target monitoring information is first extracted by filtering information related to the tower crane structural data from the synchronous monitoring model. This information includes stress data and strain data of key structural components of the tower crane, as well as data from the previously constructed structural health status mapping model. Stress and strain data directly reflect the real-time stress situation of the structure, while the structural health status mapping model data contains the correlation characteristics and health baseline information of various parts of the structure. Together, they form the core data foundation for subsequent analysis, ensuring that the analysis can cover the real-time state and historical health patterns of the structure. Next, the matching preset analysis model is called, and based on the monitoring requirements of the tower crane structural data, the anomaly detection model and fatigue life prediction model are determined as the core analysis tools. The anomaly detection model focuses on identifying whether there are any immediate anomalies in the current structure, while the fatigue life prediction model focuses on assessing the life status of the structure after long-term use. The two work together to cover both real-time safety risks and long-term operation and maintenance needs, forming a comprehensive analysis system for the tower crane structure. Then, structural anomaly identification is carried out through the anomaly detection model, a process that is carried out in two stages. The first stage employs 1D-CNN to extract local features from the temporal structural data in the target monitoring information. Temporal structural data consists of stress and strain data that dynamically change over time. 1D-CNN can capture subtle fluctuations in this data over time, such as sudden changes in stress values ​​or abnormal inflection points in strain curves. These features are often early signals of structural anomalies. The second stage uses a two-stage Canopy-KMeans clustering algorithm to process the extracted local features. First, the Canopy algorithm quickly groups the feature data to identify datasets that may contain anomalous features. Then, the KMeans algorithm performs precise clustering on these datasets, grouping similar anomalous patterns together to identify the structural components with anomalies, their specific locations, and the types of anomalies, ultimately forming a complete structural anomaly detection result. Simultaneously, a fatigue life prediction model is used for life assessment. The stress time-series data from the target monitoring information is input into an LSTM. LSTM has the ability to process long-term time-series data and can learn the stress change patterns of the tower crane under cyclic loads, thereby understanding the process of structural fatigue accumulation. For example, the correlation between the accumulation rate of fatigue damage and the number of stress cycles and the degree of fatigue under different load intensities. Based on this, by combining a BP neural network and a finite element sample analysis algorithm, the BP neural network learns the mapping relationship between structural parameters (such as material properties and component dimensions) and fatigue life in the finite element samples, and supplements and calibrates the fatigue accumulation law learned by the LSTM. Finally, it accurately predicts the remaining fatigue life of the structure, and analyzes and outputs the key factors affecting fatigue accumulation, such as frequent heavy-load operations and continuous stress in stress concentration areas.Finally, the results are integrated, summarizing and organizing the structural anomaly detection results, remaining fatigue life prediction values, and corresponding fatigue accumulation factors. Correlation analysis is performed on the various results, such as determining whether structural anomalies accelerate fatigue accumulation and whether fatigue accumulation factors are related to the current anomaly type, ensuring that all information corroborates each other and is logically consistent. Ultimately, this results in a dynamic analysis of the tower crane's structural data, comprehensively reflecting the real-time safety status and long-term life trend of the tower crane structure.

[0064] Optionally, if a sudden increase in stress is detected in the tower crane structural data, the secondary verification of the anomaly detection model is triggered first, and dynamic analysis results for the tower crane structural data are regenerated. Specifically, when a sudden increase in stress is detected in the tower crane structural data, the first step is a real-time monitoring and triggering mechanism for the stress increase. The system continuously tracks the stress parameters in the tower crane structural data dynamically, and determines whether a sudden increase has occurred by using a preset stress change rate threshold (such as stress increase exceeding a certain standard value per unit time). Once a stress increase that meets the conditions is detected, the system immediately activates a priority scheduling mechanism, suspends or delays other non-critical analysis tasks, and prioritizes the allocation of computing resources to the secondary verification of the anomaly detection model, ensuring that this process is not interfered with by other tasks and responds quickly to potential structural safety risks. Then, the detailed analysis phase of the secondary verification begins. The anomaly detection model re-extracts the full time-series data of the stress increase period from the synchronous monitoring model, including the stress baseline before the increase, the real-time stress value during the increase, and the stress stability state after the increase. The data time span is expanded compared to the initial detection to capture a more complete stress change trend. During the feature extraction stage, 1D-CNN adjusts the convolution kernel parameters and uses a smaller time step to extract local features from the data during periods of sudden stress increases. It focuses on capturing detailed features such as the magnitude, duration, and frequency of stress increases, avoiding misjudgments due to feature omissions. In the anomaly pattern recognition stage, the Canopy-KMeans two-level clustering algorithm tightens the cluster boundaries, lowers the threshold for anomaly pattern judgment, and incorporates a feature library of historical stress increase cases for comparison. This distinguishes between genuine structural anomalies (such as stress concentration caused by component deformation) and interfering factors (such as instantaneous sensor malfunctions), outputting more accurate anomaly localization and type judgment. Subsequently, the verification results are compared and corrected. The structural anomaly detection results obtained from the second verification are compared with the first detection results to analyze the differences in anomaly location, type, and severity. If the difference stems from incomplete data or parameter settings, the second verification result prevails. If the difference is small, the final anomaly information is determined by combining the commonalities between the two results. Simultaneously, the factor of sudden stress increase is incorporated into the supplementary analysis of fatigue life prediction to assess whether the sudden increase accelerates fatigue accumulation, adjust the remaining fatigue life prediction value and corresponding accumulation factors, and ensure that the long-term life assessment can reflect the impact of immediate stress anomalies. Finally, the dynamic analysis results are updated and synchronized. The corrected structural anomaly detection results and adjusted remaining fatigue life data are integrated to form new dynamic analysis results for tower crane structural data. This result is fed back to the synchronous monitoring model in real time, updating the safety status indicators of corresponding structural components in the model (such as enhancing the color warning intensity of the heat map), and triggering the system's graded early warning (such as audible and visual prompts or maintenance instruction pushes), ensuring that relevant personnel are promptly aware of the safety risks brought about by sudden stress increases, and realizing a closed loop from anomaly monitoring to decision support.

[0065] Understandably, during the feature extraction stage, 1D-CNN is specifically adjusted to capture the characteristics of data during periods of sudden stress increases. By optimizing the convolutional kernel parameters and time step, it achieves accurate capture of detailed features. The specific process is as follows: First, the convolutional kernel parameters are adjusted. Originally used for feature extraction of conventional structural data, 1D-CNN's convolutional kernel size and stride settings focus more on capturing long-term trends. However, for short-term, high-frequency changes like sudden stress increases, the algorithm reduces the temporal dimension of the convolutional kernel. For example, the convolutional kernel originally covering 10 time points is adjusted to cover 3 to 5 time points, while the sliding stride is reduced, shortening the sliding interval of the convolutional kernel on the time-series data. This adjustment allows the convolutional operation to scan each time node of the stress increase period more densely, avoiding skipping key time points such as the initial moment of the increase and the peak point due to an excessively large stride, ensuring that no subtle stress changes are missed. Second, local features are extracted specifically. For the amplitude features of stress increases, 1D-CNN uses multiple parallel groups of convolutional kernels to capture changes in different amplitude ranges. Small-scale convolutional kernels focus on the stress rise slope in the initial stage of a stress spike, calculating the stress difference between adjacent time points to quantify the steepness of the spike. Slightly larger convolutional kernels cover the peak stress during the spike, extracting amplitude features exceeding the normal range by comparing them with the maximum stress value under historical normal operating conditions. For duration features, the algorithm uses multiple convolutional layers. Shallow convolutional layers capture the stress state within a single time window, while deeper convolutional layers fuse the temporal correlation between adjacent windows, determining the duration of the spike by whether the stress values ​​remain high across multiple consecutive windows. If the stress values ​​in multiple consecutive windows exceed a threshold, it is considered a sustained spike; otherwise, it is classified as a transient fluctuation. For fluctuation frequency features, 1D-CNN introduces convolutional kernels with different frequency responses. High-frequency response convolutional kernels capture rapid oscillations during stress spikes (such as small high-frequency fluctuations after a spike), while low-frequency response kernels capture features related to sensor malfunctions or structural resonance. The low-frequency response convolution kernel focuses on the stable state of stress after a sudden increase, judging whether it exhibits regular fluctuations (such as periodic fluctuations with load changes) to distinguish between normal structural stress adjustment and abnormal vibration. Through this refined parameter adjustment and feature extraction strategy, 1D-CNN can accurately distinguish between real stress surges (such as continuous and regular surges caused by structural deformation) and interference factors (such as short-term and irregular fluctuations caused by instantaneous sensor failures), thus avoiding misjudging normal conditions as abnormalities or missing true structural anomaly signals due to feature capture failures.

[0066] In the anomaly pattern recognition stage, the application of the Canopy-KMeans two-level clustering algorithm requires hierarchical processing and parameter optimization to achieve accurate identification of anomaly patterns. The specific process is as follows: First, the preliminary clustering of the Canopy algorithm is initiated. Based on the local feature data extracted during periods of sudden stress increase, the algorithm sets a relatively loose initial distance threshold and quickly scans all feature samples, initially grouping samples with similar features into several loose candidate clusters. This step does not pursue clustering accuracy but quickly filters out samples that clearly do not belong to anomaly patterns (such as normal features during periods of stable stress), narrowing the scope of subsequent fine analysis, while providing initial cluster centers for the KMeans algorithm. Next, the fine clustering stage of the KMeans algorithm begins. For the candidate clusters output by the Canopy algorithm, KMeans adopts a more stringent distance calculation method (such as Euclidean distance or cosine similarity), iteratively optimizing the cluster centers to minimize the feature differences of samples within the same cluster and maximize the differences between different clusters. At this point, the algorithm actively tightens the cluster boundaries. This involves reducing the maximum allowable distance between samples within the same cluster, ensuring that similar but subtly different features that might otherwise be grouped into the same cluster are distinguished into different clusters. For example, it clearly separates features of sustained stress surges from those of brief stress fluctuations, providing a more refined clustering basis for subsequent anomaly detection. Simultaneously, the algorithm lowers the threshold for anomaly pattern detection. The feature deviation standard originally used to determine whether a cluster is an anomaly pattern is lowered, allowing clusters with features that deviate slightly from the normal baseline to be included in the anomaly candidate range. This adjustment makes the algorithm more sensitive to potential anomalies, avoiding the omission of early anomaly signals that might be hidden in stress surges due to excessively high thresholds, especially for stress anomaly changes with small amplitude but significant trends. Based on this, the algorithm introduces a feature library of historical stress surge cases for comparison. The feature library stores feature vectors of past real structural anomalies (such as stress concentration caused by component deformation) and various interference factors (such as instantaneous sensor failures), including the amplitude variation curve of stress surges, duration distribution, and accompanying strain / vibration characteristics. The algorithm calculates the similarity between the features of the current clustered candidate clusters and the features of each category in the feature library. If the feature similarity between a candidate cluster and a real structural anomaly case is higher than a set standard, and the features of that category all correspond to actual structural damage in the matched historical cases, then it is determined to be a real anomaly. If the feature similarity with sensor failure cases is higher (e.g., sudden and irregular increase followed by instantaneous recovery, with no corresponding strain change), then it is marked as an interfering factor. Finally, combining the clustering results with historical comparison conclusions, the algorithm outputs accurate anomaly location and type judgment: clearly identifying the specific structural component where the anomaly occurred (e.g., a specific connection point of a standard section of a tower), distinguishing the anomaly type (e.g., stress concentration, component deformation, or sensor interference), and marking the confidence level of the anomaly (e.g., the degree of matching with historical real anomalies), providing a reliable basis for subsequent structural safety assessments.

[0067] As an optional embodiment, in step S104, a preset analysis model matching the target display dimension is invoked, target monitoring information is extracted from the synchronous monitoring model, and the target monitoring information is subjected to tower crane dynamic analysis through the preset analysis model to obtain the tower crane dynamic analysis result. This includes: when the target display dimension is the tower crane movement trajectory, the tower crane movement trajectory data is extracted from the synchronous monitoring model as target monitoring information, the target monitoring information including slewing angle, trolley displacement, and movement trajectory data; a tower crane operation risk warning model is invoked as the matching preset analysis model; in the tower crane operation risk warning model, a fuzzy neural network is used to fuse trolley position and movement trajectory prediction data, combined with the A* algorithm to generate a safe trajectory, and PointLSTM is used to predict the trajectory of the point cloud data associated with the tower group to obtain the collision risk level and warning threshold; the safe trajectory, collision risk level, and warning threshold are integrated as the dynamic analysis result for the tower crane movement trajectory.

[0068] It can be explained that in step S104 above, when the target display dimension is the tower crane's movement trajectory, the target monitoring information is first extracted. The full amount of data related to the tower crane's movement trajectory is selected and obtained from the synchronous monitoring model as the target monitoring information. This data includes the real-time slewing angle of the tower crane's slewing mechanism, the trolley displacement of the luffing mechanism, and the complete movement trajectory sequence formed by the changes of these two parameters over time. For example, the slewing angle reflects the horizontal rotation position of the tower crane's boom, and the trolley displacement reflects the forward and backward movement distance of the trolley on the boom. The combination of these two constitutes the planar position information of the tower crane's operation, while the movement trajectory sequence records the position change process at different time points, providing a time-series data basis for subsequent trajectory prediction and risk assessment. Next, a matching preset analysis model is called, and based on the monitoring requirements of the tower crane's movement trajectory, the tower crane operation risk warning model is determined as the core analysis tool. This model focuses on identifying potential collision risks during the tower crane's movement (such as collisions with other tower cranes or buildings) and generating safe paths to avoid risks, making it a key model for ensuring the safety of tower crane operations. Subsequently, the model enters a multi-algorithm collaborative analysis phase, completing risk assessment and safe trajectory generation in three steps. The first step is location data fusion based on a fuzzy neural network: fuzzy neural networks can handle uncertainties in the data (such as small errors in trolley displacement measurement and instantaneous fluctuations in rotation angle), fusing the trolley's real-time position from target monitoring information with the predicted motion trajectory data obtained through prior calculations. The weights of the two types of data are dynamically adjusted using fuzzy rules; for example, increasing the weight of real-time location data when its accuracy is high, and enhancing the synergy between the predicted and real-time data when the deviation is small, ultimately outputting more stable and reliable current and short-term predicted position information. The second step is safe trajectory generation based on the A* algorithm. The A* algorithm starts with the fused location data, combines environmental constraints of the work site (such as pre-recorded building locations and prohibited work area boundaries), and real-time location information from other tower cranes, iteratively searching for the optimal path by calculating the comprehensive cost of the path length from the current position to the target position and the degree of risk along the path. During the search process, the algorithm avoids all areas with potential collision risks, ultimately generating a safe motion trajectory that meets operational requirements while mitigating immediate risks. The third step involves predicting the trajectory of multiple tower cranes and determining the risk level based on PointLSTM. PointLSTM is capable of processing 3D point cloud data and temporal information, and can receive point cloud data of all tower cranes in a multi-tower operation scenario (including the structural outline and real-time position coordinates of each tower crane). By learning the temporal patterns of historical multi-tower movement, it predicts the movement trajectory of other tower cranes in the future. The safe trajectory of the current tower crane is compared with the predicted trajectory of other tower cranes, and the probability and time window of trajectory intersection are calculated. If the intersection probability is higher than a set value, it is judged as high risk; if there is a potential intersection but there is sufficient time to avoid it, it is judged as medium risk; if the trajectories do not overlap, it is judged as low risk.Simultaneously, corresponding early warning thresholds are determined based on the risk level (e.g., an emergency deceleration threshold for high risk, and an early warning threshold for medium risk). Finally, the analysis results are integrated, summarizing the safety trajectory generated by the A* algorithm, the collision risk level determined by PointLSTM, and the corresponding early warning thresholds. The safety trajectory clarifies the optimal path for the tower crane's subsequent actions, the risk level intuitively reflects the current safety status of the operation, and the early warning threshold provides a quantitative standard for triggering safety interventions (such as automatic deceleration or stopping the action). The combination of these three elements forms a dynamic analysis result for the tower crane's movement trajectory, providing action guidance for operators and a basis for the system's automatic safety control.

[0069] Optionally, when a sudden change in wind speed in environmental monitoring data exceeds the set wind speed limit, the calculation weight of the tower crane operation risk warning model is increased, and the collision risk level and warning threshold are adjusted based on the updated calculation weight. Specifically, the above steps begin with real-time monitoring and weight adjustment triggering of sudden changes in wind speed. The system continuously collects real-time wind speed data from environmental monitoring data and calculates the rate of change of wind speed per unit time through a sliding window. When the rate of change exceeds the preset sudden change judgment threshold and the instantaneous wind speed exceeds the set wind speed limit, the system immediately triggers the instruction to adjust the model calculation weight. At this time, the system pauses the model's current regular calculation process and prioritizes allocating computing resources to the weight update module to ensure that the adjustment action responds quickly to wind speed anomalies and avoids delays in risk assessment. Next, the calculation weight of the tower crane operation risk warning model is increased. The weights of calculation modules related to the impact of wind load are specifically increased. These modules include the wind-induced trajectory deviation calculation module, the mechanism action error compensation module under wind load, and the tower crane trajectory uncertainty prediction module. After the weight of the wind-induced trajectory deviation calculation module is increased, it will focus more on analyzing the actual impact of sudden changes in wind speed on the tower crane's movement trajectory. For example, strong winds can increase the swing amplitude of the crane boom, leading to a deviation between the actual displacement of the trolley and the theoretical trajectory. The model will increase the proportion of this deviation in the trajectory fusion calculation. After increasing the weight of the mechanism motion error compensation module under wind load, it will strengthen the calculation of motion delays and speed fluctuations of the slewing mechanism and luffing mechanism caused by increased wind resistance, ensuring that the predicted trajectory reflects the actual motion state of the mechanism under strong wind conditions. After increasing the weight of the tower crane trajectory uncertainty prediction module, it will expand the prediction range of trajectory deviations caused by sudden changes in wind speed for other tower cranes, avoiding the omission of collision risks due to underestimating the wind-induced deviations of adjacent tower cranes. Subsequently, the risk level is recalculated based on the updated weights. The model will reintegrate target monitoring information (slewing angle, trolley displacement, motion trajectory data) and sudden wind speed data based on the increased weights, and conduct multi-algorithm collaborative analysis. When fusing trolley position and trajectory prediction data, the fuzzy neural network will use the trajectory deviation caused by sudden changes in wind speed as the core input parameter, increasing the weight of this parameter in the fusion result, so that the output predicted position is closer to the actual position under strong wind conditions. When generating safe trajectories, the A* algorithm expands the obstacle avoidance distance based on wind-induced trajectory offset. For example, the originally reserved safety distance is increased proportionally according to wind speed, ensuring that even if the trajectory deviates due to wind, it can still avoid obstacles or other tower cranes. When predicting the point cloud data of a group of tower cranes, PointLSTM incorporates trajectory uncertainty factors caused by sudden changes in wind speed, expanding the prediction range of future trajectories of other tower cranes and making collision risk assessment more comprehensive. Finally, the collision risk level and warning threshold are adjusted. The model adjusts the collision risk level based on the recalculated trajectory deviation, safety distance, and probability of overlap between tower crane trajectories: a trajectory originally judged as low-risk will be upgraded to medium-risk if the probability of overlap with other tower crane trajectories increases due to sudden changes in wind speed.If a trajectory initially classified as medium-risk overlaps with a new risk threshold, it will be upgraded to high-risk. Simultaneously, the warning threshold will be lowered accordingly. For example, the trajectory overlap probability threshold triggering a medium-risk warning will be lowered from a previous value, allowing the system to issue warnings earlier when the risk level is low. The threshold for triggering high-risk emergency interventions (such as automatic deceleration or stopping) will also be moved forward, ensuring more time to avoid collisions. Ultimately, this results in a risk assessment adapted to scenarios of sudden wind speed changes, ensuring the safe operation of tower cranes in adverse wind conditions.

[0070] As an optional embodiment, in step S104, a preset analysis model matching the target display dimension is invoked, target monitoring information is extracted from the synchronous monitoring model, and the target monitoring information is subjected to tower crane dynamic analysis through the preset analysis model to obtain tower crane dynamic analysis results. This includes: when the target display dimension is tower crane image data, tower crane image data is extracted from the synchronous monitoring model as target monitoring information, the target monitoring information including obstacle and load status data; a tower crane operation risk warning model is invoked as the matching preset analysis model; in the tower crane operation risk warning model, the obstacle volume, load position, and load volume extracted from the target monitoring information are fused through a fuzzy neural network, and an avoidance trajectory is generated by combining the A* algorithm, outputting the obstacle collision risk level and load status assessment results; based on the real-time monitored obstacle distance, load sway amplitude, rotation angle, and trolley displacement, reinforcement learning is used to adjust the target weights of different obstacle collision risk levels, prioritizing the response to the dynamic analysis needs of high-risk targets, and forming dynamic analysis results for tower crane image data.

[0071] Specifically, the above steps begin with the extraction of target monitoring information. Tower crane image data is selected from the synchronous monitoring model as the core analysis object. This data focuses on the status of obstacles and suspended loads identified in the images. Obstacle information includes their size and spatial location (such as the horizontal and vertical distance from the tower crane boom), while the suspended load status data covers the real-time position (coordinates relative to the hook and the ground), volume, binding status, and swaying. This data directly reflects potential collision hazards and the safety status of the suspended load in the operating scenario, forming the basis for subsequent risk assessment. Next, the tower crane operation risk warning model is used as the core analysis tool. This model is specifically designed for the visual characteristics of image data, addressing obstacle avoidance and suspended load safety assessment needs. It achieves risk quantification and trajectory optimization through multi-algorithm collaboration. The analysis process within the model proceeds in two steps. The first step involves data fusion and trajectory generation based on fuzzy neural networks. Fuzzy neural networks can handle uncertainties in image data caused by perspective deviations and occlusions (such as minor errors in obstacle volume estimation and visual offsets of suspended load positions), fusing the extracted obstacle volume, suspended load position, and suspended load volume. The weights of each parameter are dynamically adjusted using preset fuzzy rules. For example, when the obstacle is large, its proportion in the collision risk calculation is increased; when the load exceeds the normal range, the reliability of its position data is strengthened. The fused data is input into the A* algorithm, which combines the spatial distribution of obstacles with the motion constraints of the load (such as the safe radius when the load sways) to plan an avoidance trajectory that can both avoid obstacles and adapt to the state of the load. At the same time, the model calculates the collision risk level based on the distance between the obstacle and the tower crane and the relative motion trend (e.g., very close and rapidly approaching is high risk, far and stationary is low risk), and outputs the load state assessment result (e.g., stable, slight swaying, severe tilting) by analyzing the swaying amplitude and the integrity of the harness. In the second step, the target weights are dynamically adjusted based on reinforcement learning. The model continuously receives real-time monitoring data, including the real-time distance between the obstacle and the tower crane, the swaying amplitude of the load, the rotation angle of the tower crane, and the displacement of the trolley, and uses this data as the state input for reinforcement learning. The reinforcement learning agent dynamically adjusts the target weights for different obstacle collision risk levels by learning from historical interaction data (such as past incidents of untimely responses to high-risk obstacles or safety events caused by violent shaking of suspended objects). For high-risk targets (such as nearby, large obstacles or violently shaking suspended objects), their priority weight in the analysis queue is increased, ensuring that the model prioritizes risk assessment and trajectory optimization for these targets. For low-risk targets, their weights are appropriately reduced to decrease computational resource consumption. This adjustment allows the model to flexibly allocate analysis resources based on the urgency of risks in the real-time scenario, prioritizing the dynamic analysis needs of high-risk targets.Finally, the analysis results are integrated, summarizing the avoidance trajectory generated by the A* algorithm, the obstacle collision risk level, the load status assessment results, and the target priority information adjusted by reinforcement learning, to form a dynamic analysis result for tower crane image data. This result not only clarifies the specific path for obstacle avoidance and quantifies the current risk status, but also guides the system to prioritize high-risk scenarios, providing operators with intuitive safety guidance and decision-making basis for tower crane automatic control.

[0072] As an optional embodiment, in step S105, updating the tower crane dynamic analysis results to the synchronous monitoring model to update the model display information matching the target display dimension in the synchronous monitoring model can be achieved as follows: First, the dynamic analysis results are classified and matched with model modules. The system first identifies the target display dimension corresponding to the tower crane dynamic analysis results. If the result originates from the analysis of tower crane structural data (such as structural anomaly detection results, remaining fatigue life data), the physical model and structural safety visualization module of the synchronous monitoring model are matched. If the result originates from the analysis of tower crane motion trajectory (such as safety trajectory, collision risk level), the motion simulation module and risk warning identification module of the model are matched. If the result originates from the analysis of tower crane image data (such as avoidance trajectory, load status assessment), the visual fusion module and target annotation module of the model are matched. Through this classification and matching, the update position of each type of analysis result in the synchronous monitoring model is clarified, avoiding information mismatch. Then, specific update operations are performed according to the dimensions.

[0073] For tower crane structural data analysis results, the specific location and type of anomalies identified in the structural anomaly detection results are synchronized to the corresponding component parameters in the physical model. For example, the mechanical calculation coefficients of the anomalous components in the physical model can be adjusted (e.g., lowering the stress bearing threshold) to simulate the stress changes in a real structure. Simultaneously, the information displayed in the structural safety visualization module is updated, such as enhancing the color warning of the heatmap for anomalous areas (e.g., upgrading from yellow to red), and overlaying text annotations (e.g., "Surging stress area, remaining lifespan XX days") at the corresponding locations in the model, visually presenting changes in the structural safety status.

[0074] Based on the tower crane motion trajectory analysis results, the safety trajectory data generated by the A* algorithm replaces the original trajectory prediction path in the motion simulation module of the synchronous monitoring model, ensuring that the motion paths of virtual tower crane components (such as the boom and luffing trolley) in the model are completely consistent with the safety trajectory. Simultaneously, the collision risk level and warning threshold are updated in the risk warning indicator module. If the risk level is high, a red dynamic warning box (covering the risk area) is triggered in the model, and a prompt "Emergency deceleration threshold triggered; immediate adjustment of actions recommended" is displayed in the corner of the interface. If the risk level is medium, a yellow warning box is triggered, prompting "Pay attention to adjacent tower crane trajectories and allow time for avoidance."

[0075] For tower crane image data analysis results, the avoidance trajectories generated by the fuzzy neural network and the A* algorithm are superimposed onto the visual fusion module of the synchronous monitoring model. In the virtual viewpoint corresponding to the image data (such as the viewpoint of the camera at the end of the boom), the avoidance path is marked with a dynamic dashed line to clearly guide the direction of avoiding obstacles; at the same time, the status assessment result of the hoisted object (such as "the hoisted object is slightly swaying, and the binding is normal") is updated to the target annotation module, and status icons (such as green stable icon and yellow swaying icon) are superimposed on the virtual position of the hoisted object in the model, and the numerical annotation of the sway amplitude of the hoisted object is updated in real time.

[0076] Finally, a consistency check and interface synchronization are performed after the update. Key parameters of the updated synchronized monitoring model (such as stress values ​​of structural components, coordinates of motion trajectories, and positions of image targets) are compared with the core data of the dynamic analysis results to ensure no discrepancies. If minor conflicts exist (such as path deviations between the virtual trajectory and the analysis results), the model parameters are automatically fine-tuned until consistency is achieved. After successful verification, the updated model display information is pushed to the user interface in real time, overwriting the original display content and ensuring that the model status seen by the user is completely synchronized with the latest dynamic analysis results.

[0077] This application's embodiments achieve full-dimensional, high-precision perception of tower crane operating status and customized display based on viewing commands. It solves the technical problems of fragmented data collection, lagging analysis methods, monotonous visualization presentation, and delayed response in related technologies. Furthermore, it improves the efficiency and accuracy of operation and maintenance decisions through dynamic adaptation and intelligent analysis, reduces safety risks in tower crane operation, and reduces manual operation and maintenance costs.

[0078] After introducing the methods of exemplary embodiments of this application, the following references are made. Figure 2This application describes an exemplary embodiment of a tower crane multidimensional data dynamic analysis and visualization system. A data acquisition module is used to acquire multidimensional monitoring data of the tower crane; the multidimensional monitoring data includes at least: tower crane structural data, tower crane movement trajectory, environmental monitoring data, and tower crane image data. A modeling module is used to establish a synchronous monitoring model of the tower crane based on the multidimensional monitoring data using digital twin technology, and transmit it to the user interface. The user interface is used to display the synchronous monitoring model; the model display information corresponding to the folded dimensions is hidden in the synchronous monitoring model. A selection module is used to receive user viewing instructions for the synchronous monitoring model, match target display dimensions for the user based on the viewing instructions, and synchronously set folded dimensions that do not match the viewing instructions. The analysis module is used to call a preset analysis model that matches the target display dimension, extract target monitoring information from the synchronous monitoring model, and perform tower crane dynamic analysis on the target monitoring information through the preset analysis model to obtain tower crane dynamic analysis results. The preset analysis model includes one or a combination of the following models: anomaly detection model, fatigue life prediction model, and tower crane operation risk early warning model. The tower crane dynamic analysis results are then updated in the synchronous monitoring model to update the model display information in the synchronous monitoring model that matches the target display dimension. The above system can implement the steps described in the above method implementation, and the specific implementation methods of each step will not be repeated here.

[0079] After introducing the methods and systems of the exemplary embodiments of this application, a terminal device of the exemplary embodiments of this application will be described next. The terminal device can implement the steps described in the above method embodiments, and the specific implementation of each step will not be repeated here.

[0080] It should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Therefore, the protection scope of this application should be determined by the scope of the claims.

Claims

1. A method for dynamic analysis and visualization of multidimensional data of tower cranes, characterized in that, The method includes: Acquire multi-dimensional monitoring data of the tower crane; the multi-dimensional monitoring data includes at least: tower crane structural data, tower crane movement trajectory, environmental monitoring data, and tower crane image data; Based on the multidimensional monitoring data, a synchronous monitoring model of the tower crane is established using digital twin technology, and the synchronous monitoring model is displayed synchronously in the user interface. This includes: performing spatiotemporal alignment and denoising on the tower crane structural data, tower crane motion trajectory, environmental monitoring data, and tower crane image data from the multidimensional monitoring data using Kalman filtering and particle filtering algorithms to generate a fused data frame; constructing a digital twin model containing a geometric model, a physical model, and an environmental model based on the fused data frame; establishing a structural health status mapping model using a graph convolutional network based on the geometric and physical models; calculating the reconstruction error through the structural health status mapping model, detecting whether there are structural anomalies in the tower crane, and outputting the result. The system includes a tower crane structural safety assessment result containing reconstruction errors and structural anomaly detection results; based on a digital twin model as the basic framework, keyframe interpolation is performed on the dynamic attribute data in the fused data frame in the GPU shader, and the corresponding abnormal change area in the digital twin model is located based on the tower crane structural safety assessment result. The interpolation calculation result is updated through an event-driven incremental rendering algorithm to obtain a real-time rendering load; the real-time rendering load is displayed in the user interface; combining different data types in the fused data frame, the non-dynamic attribute data in the fused data frame is adapted to a visualization form, and synchronously rendered to the user interface according to the adapted visualization form to obtain the synchronous monitoring model; The system receives a user's viewing instruction for the synchronous monitoring model, matches a target display dimension for the user based on the viewing instruction, synchronously sets a collapsed dimension that does not match the viewing instruction, and hides the model display information corresponding to the collapsed dimension in the synchronous monitoring model. The steps of receiving the user's viewing instruction for the synchronous monitoring model, matching a target display dimension for the user based on the viewing instruction, and synchronously setting a collapsed dimension that does not match the viewing instruction include: receiving the user-initiated viewing instruction through a user interface, wherein the input form of the viewing instruction includes manual text input, voice commands, and / or gesture operations; inputting the viewing instruction into a Seq2Seq-based dynamic semantic parser, parsing the viewing instruction into a structured query to obtain key instruction information, and establishing a mapping relationship between the key instruction information and each monitoring data category in the multidimensional monitoring data. The system evaluates the correlation between each monitoring data category and the key information of the instruction using a graph attention network based on the mapping relationship; it selects monitoring data categories whose correlation with the key information of the instruction is higher than a correlation threshold as target display dimensions, and marks monitoring data categories whose correlation with the key information of the instruction is not higher than a correlation threshold as folded dimensions; it uses a reinforcement learning agent to dynamically adjust the correlation threshold based on the context information of the viewing instruction and historical interaction feedback information; it combines a historical interaction database and uses a K-means clustering algorithm to perform cluster analysis on the user's past viewing dimension preferences; it optimizes and adjusts the target display dimensions and folded dimensions based on the cluster analysis results, retaining monitoring data categories whose user attention frequency reaches a preset condition as the final output target display dimensions, and using monitoring data categories whose user attention frequency does not reach a preset condition as folded dimensions; A preset analysis model matching the target display dimension is invoked, target monitoring information is extracted from the synchronous monitoring model, and tower crane dynamic analysis is performed on the target monitoring information through the preset analysis model to obtain tower crane dynamic analysis results; wherein the preset analysis model includes one or a combination of the following models: anomaly detection model, fatigue life prediction model, and tower crane operation risk early warning model; The tower crane dynamic analysis results are updated to the synchronous monitoring model to update the model display information that matches the target display dimension in the synchronous monitoring model.

2. The method for dynamic analysis and visualization of multidimensional data of tower cranes according to claim 1, characterized in that, The construction of a digital twin model comprising a geometric model, a physical model, and an environmental model based on the fused data frame includes: The tower crane structural dimensions and component location data from the fused data frame are used to construct a geometric model using a 3D modeling tool. The structural mechanical behavior of the tower and boom in the tower crane is simulated by finite element analysis combined with the original data of structural stress and strain in the fused data frame. Multibody dynamics is used in combination with the motion trajectory parameters in the fused data frame to simulate the real-time motion trajectory of each moving part in the tower crane. The physical model is obtained by fusing the structural mechanical behavior and the real-time motion trajectory. An environmental model is obtained by integrating the real-time environmental data from the fused data frame. By integrating geometric models, physical models, and environmental models, a digital twin model is formed.

3. The method for dynamic analysis and visualization of multidimensional data of tower cranes according to claim 1, characterized in that, The process of combining different data types in the fused data frame to adapt the non-dynamic attribute data in the fused data frame to a visualization format includes: For the point cloud data in the fused data frame, the PointLSTM model is called to predict the motion trajectory, and the motion trajectory prediction result is embedded into the synchronous monitoring model. The tower crane structural safety assessment results output by the graph convolutional network are superimposed onto the corresponding structural components in the synchronous monitoring model in the form of a heat map; For the environmental monitoring data in the fused data frame, a spatiotemporal grid method is used to mark the synchronous monitoring model, and dynamic annotations are superimposed. For the image data in the fused data frame, YOLO and Faster R-CNN are used to detect obstacles and suspended objects, and the target detection results are embedded into the corresponding viewpoint in the synchronous monitoring model.

4. The method for dynamic analysis and visualization of multidimensional data of tower cranes according to claim 1, characterized in that, The process involves calling a preset analysis model that matches the target display dimension, extracting target monitoring information from the synchronous monitoring model, and performing tower crane dynamic analysis on the target monitoring information using the preset analysis model to obtain tower crane dynamic analysis results, including: When the data type corresponding to the target display dimension is tower crane structure data, the tower crane structure data is extracted from the synchronous monitoring model as target monitoring information. The target monitoring information includes stress and strain data and structural health status mapping model data. The anomaly detection model and fatigue life prediction model are used as the preset analysis models for matching; Using an anomaly detection model, 1D-CNN is used to extract local features from the temporal structure data in the target monitoring information, and Canopy-KMeans two-level clustering algorithm is used to perform tower crane anomaly pattern recognition on the local feature extraction results, so as to obtain structural anomaly detection results including the abnormal structure, its location, and the type of anomaly. By using the fatigue life prediction model, the stress time series data in the target monitoring information is input into LSTM to learn the fatigue accumulation law of the tower crane under cyclic load. The remaining fatigue life prediction value and the corresponding fatigue accumulation factors are predicted by combining the BP neural network and the finite element sample analysis algorithm. By integrating structural anomaly detection results, predicted remaining fatigue life, and corresponding fatigue accumulation factors, dynamic analysis results for tower crane structural data are formed.

5. The method for dynamic analysis and visualization of multidimensional data of tower cranes according to claim 1, characterized in that, The process involves calling a preset analysis model that matches the target display dimension, extracting target monitoring information from the synchronous monitoring model, and performing tower crane dynamic analysis on the target monitoring information using the preset analysis model to obtain tower crane dynamic analysis results, including: When the target display dimension is the tower crane's motion trajectory, the tower crane's motion trajectory data is extracted from the synchronous monitoring model as target monitoring information. The target monitoring information includes rotation angle, trolley displacement, and motion trajectory data. The tower crane operation risk warning model is used as the preset analysis model for matching. In the tower crane operation risk warning model, a fuzzy neural network is used to fuse the position and trajectory data of the trolley, and the A* algorithm is combined to generate a safe trajectory. PointLSTM is used to predict the trajectory of the point cloud data associated with the tower group to obtain the collision risk level and warning threshold. The safety trajectory, collision risk level, and early warning threshold are integrated as dynamic analysis results for the tower crane's movement trajectory.

6. The method for dynamic analysis and visualization of multidimensional data of tower cranes according to claim 1, characterized in that, The process involves calling a preset analysis model that matches the target display dimension, extracting target monitoring information from the synchronous monitoring model, and performing tower crane dynamic analysis on the target monitoring information using the preset analysis model to obtain tower crane dynamic analysis results, including: When the target display dimension is tower crane image data, the tower crane image data is extracted from the synchronous monitoring model as target monitoring information, and the target monitoring information includes obstacle and load status data. The tower crane operation risk warning model is called as the preset analysis model for matching; in the tower crane operation risk warning model, the obstacle volume, load position and load volume extracted from the target monitoring information are fused by the fuzzy neural network, and the avoidance trajectory is generated by combining the A* algorithm, and the obstacle collision risk level and load status assessment results are output. Based on real-time monitoring of obstacle distance, load sway amplitude, rotation angle, and trolley displacement, reinforcement learning is used to adjust the target weights for different obstacle collision risk levels, prioritizing responses to the dynamic analysis needs of high-risk targets, and generating dynamic analysis results for tower crane image data.

7. A multi-dimensional data dynamic analysis and visualization system for tower cranes, characterized in that, The system includes the following modules: The data acquisition module is used to acquire multidimensional monitoring data of the tower crane; The multidimensional monitoring data includes at least: tower crane structural data, tower crane movement trajectory, environmental monitoring data, and tower crane image data; The modeling module is used to establish a synchronous monitoring model of the tower crane based on the multidimensional monitoring data using digital twin technology, and then transmit it to the user interface; When the modeling module establishes a synchronous monitoring model of the tower crane using digital twin technology based on the multidimensional monitoring data, it specifically performs the following: For the tower crane structural data, tower crane motion trajectory, environmental monitoring data, and tower crane image data in the multidimensional monitoring data, it uses Kalman filtering and particle filtering algorithms to perform spatiotemporal alignment and denoising to generate a fused data frame; based on the fused data frame, it constructs a digital twin model containing a geometric model, a physical model, and an environmental model; based on the geometric model and the physical model, it establishes a structural health state mapping model using a graph convolutional network; it calculates the reconstruction error through the structural health state mapping model, detects whether there are structural anomalies in the tower crane, and outputs a model containing the reconstruction error. The tower crane structural safety assessment results are compared with the structural anomaly detection results; based on the digital twin model as the basic framework, keyframe interpolation is performed on the dynamic attribute data in the fused data frame in the GPU shader, and the corresponding abnormal change area in the digital twin model is located based on the tower crane structural safety assessment results. The interpolation calculation results are updated through an event-driven incremental rendering algorithm to obtain the real-time rendering load; the real-time rendering load is displayed in the user interface; combined with the different data types in the fused data frame, the non-dynamic attribute data in the fused data frame is adapted to a visualization form, and synchronously rendered to the user interface according to the adapted visualization form to obtain the synchronous monitoring model; A user interface is used to display the synchronous monitoring model; the model display information corresponding to the folded dimensions is hidden in the synchronous monitoring model; The selection module is used to receive the user's viewing instruction for the synchronous monitoring model, match the target display dimension for the user based on the viewing instruction, and synchronously set the folded dimension that does not match the viewing instruction; The selection module is specifically used for: receiving viewing instructions initiated by the user through a user interface, wherein the input form of the viewing instructions includes manual text input, voice commands, and / or gesture operations; inputting the viewing instructions into a Seq2Seq-based dynamic semantic parser, parsing the viewing instructions into a structured query to obtain key information of the instructions, establishing a mapping relationship between the key information of the instructions and each monitoring data category in the multidimensional monitoring data; evaluating the correlation between each monitoring data category and the key information of the instructions using a graph attention network based on the mapping relationship; and filtering out monitoring data categories whose correlation with the key information of the instructions is higher than a correlation threshold. Data categories are used as the target display dimension. Monitoring data categories whose relevance to the key information of the instruction is not higher than the relevance threshold are marked as folded dimensions. A reinforcement learning agent is used to dynamically adjust the relevance threshold based on the context information of the viewing instruction and historical interaction feedback information. Combined with the historical interaction database, the K-means clustering algorithm is used to perform cluster analysis on the user's past viewing dimension preferences. The target display dimension and folded dimension are optimized and adjusted according to the cluster analysis results. Monitoring data categories whose user attention frequency reaches the preset condition are retained as the final output target display dimension, and monitoring data categories whose user attention frequency does not reach the preset condition are used as folded dimensions. The analysis module is used to call a preset analysis model that matches the target display dimension, extract target monitoring information from the synchronous monitoring model, and perform tower crane dynamic analysis on the target monitoring information through the preset analysis model to obtain tower crane dynamic analysis results; wherein the preset analysis model includes one or a combination of the following models: anomaly detection model, fatigue life prediction model, and tower crane operation risk early warning model; and update the tower crane dynamic analysis results to the synchronous monitoring model to update the model display information in the synchronous monitoring model that matches the target display dimension.

8. A terminal device, characterized in that, The terminal device is used to implement the tower crane multidimensional data dynamic analysis and visualization method as described in any one of claims 1 to 6.

Images