Bridge information intelligent management system

Abstract

The invention belongs to the technical field of bridge construction, and particularly relates to a bridging information intelligent management system which comprises a client sensing layer, a data acquisition layer, a logic processing layer, an intelligent analysis layer and an ecological safety layer. According to the intelligent management system, through an AR head-mounted display, an unmanned aerial vehicle group and a sensor network configured by a client sensing layer, bridging full-process multi-source data acquisition is realized, and construction monitoring millimeter-level precision alignment is achieved by using AR space anchoring and multi-sensor fusion technologies. The multi-source data space-time alignment middleware of the data acquisition layer ensures the accuracy and unification of the multi-modal data in the space-time dimension by means of GNSS / SLAM fusion solution, PTP protocol clock synchronization and a registration mode of combining an ICP algorithm with plane constraint.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge construction, and particularly relates to a bridge construction information intelligent management system. BACKGROUND

[0002] In the existing bridge construction and maintenance process, since bridge construction involves multiple aspects, corresponding standards and specifications need to be formulated to ensure the coordination and data exchange between various modules. In the absence of unified standards and specifications, the progress and quality of bridge construction are easily affected. Meanwhile, since a large amount of sensitive data is involved in the bridge information management system, the bridge engineering management has significant pain points in data collection, construction management, operation and maintenance collaboration, and data security, such as:

[0003] Data collection and monitoring: relying on manual measurement and single sensors, the precision is only centimeter level, multi-source data lacks spatio-temporal unified reference, there are "data islands", it is difficult to meet the high-precision requirements of high-speed railway bridges, and the precision of traditional AR technology and data fusion technology is insufficient;

[0004] Construction management: design-construction-operation process is fragmented, manual data transmission is low in efficiency, risk warning relies on manual inspection, lacks dynamic decision-making means, traditional BIM system cannot realize construction data feedback to design, and risk analysis lacks quantitative model;

[0005] Operation and maintenance management: manual inspection has high missing rate and slow response, multi-party collaboration data sharing is not safe and lacks trust mechanism, and single data model has low prediction accuracy;

[0006] Data security and collaboration: data storage is scattered, retrieval is inefficient, security protection means is backward, federal learning application is blank, and permission management is static.

[0007] In view of the above problems, the present application provides a bridge construction information intelligent management system. SUMMARY

[0008] Based on the technical problems of existing data collection precision, process collaboration efficiency, operation foresight and data security, the present application provides a bridge construction information intelligent management system.

[0009] The bridge construction information intelligent management system provided by the present application comprises a client perception layer, a data collection layer, a logic processing layer, an intelligent analysis layer and an ecological security layer. The client perception layer is configured with an AR head-mounted display, an intelligent construction terminal, a UAV group and a sensor network, is used for collecting bridge full-process multi-source data, and realizes millimeter-level precision alignment of construction monitoring through AR space anchoring and multi-sensor fusion technology, and covers beam hoisting and environment perception bridge scene.

[0010] The data collection layer deploys a multi-protocol gateway and distributed storage to clean, fuse and bridge multi-modal data, unify the construction coordinate system reference through spatio-temporal alignment algorithms and blockchain storage technology, associate process timestamps, and ensure traceability of the bridge data throughout the process.

[0011] The logic processing layer constructs a business process engine to run through the bridge design-construction-operation business, automatically trigger construction deviation warnings and process compliance checks through BIM model verification and dynamic decision algorithms, and drive closed-loop control of the bridge process.

[0012] The intelligent analysis layer integrates simulation and prediction models to mine the decision-making value of bridge data, optimize bridge resource scheduling, and improve construction safety and efficiency through finite element construction simulation and LSTM disease prediction technology.

[0013] The ecological security layer builds a zero-trust architecture and a collaboration platform to ensure bridge data security and multi-party collaboration, and realizes cross-organizational data sharing through blockchain smart contracts and dynamic permission control technology, meeting the needs of bridge compliance auditing.

[0014] Preferably, the logic processing layer includes a design collaboration module, an operation detection module, a construction management module, and a collaborative management module. The design collaboration module dynamically updates the model through a parameterized BIM engine and combines VR gesture interaction review to achieve collision detection. The operation detection module improves disease identification accuracy through a multi-modal data fusion prediction model and combines a digital twin to map the bridge state in real time. The construction management module optimizes the process through a PPO algorithm, adjusts resource priority, relies on a risk knowledge graph, and keeps the warning response time at the hour level. The collaborative management module uses a dynamic data sandbox to achieve multi-party data sharing.

[0015] Preferably, the intelligent analysis layer includes a digital twin modeling platform, cloud computing, and a model cluster. The digital twin modeling platform balances computing efficiency and accuracy through hybrid precision modeling technology, integrates ANSYS finite element kernel, and supports virtual pre-performance of the construction process. The cloud computing deploys TensorRT inference engines and handles bridge site displacement adjustment real-time control tasks through the edge layer. The cloud end executes large-scale finite element simulation and model training through Kubernetes elastic scheduling of GPU clusters. The model cluster progress prediction absorbs construction data in real time through random forest algorithms and reinforcement learning algorithms and aggregates multi-project data relying on federated learning.

[0016] Preferably, the ecological security layer includes a distributed storage, a security architecture and a federated learning ecology, the distributed storage adopts a hybrid architecture, a PostgreSQL storage structured work order / log HDFS storage unstructured point cloud image, a TimescaleDB storage time series sensor data, guaranteeing efficient management of multiple types of data; the security architecture dynamically adjusts data access permissions through triple verification of device identity, space-time position and biological characteristics, and core data is stored by AES-256 encryption; the federated learning ecology aggregates multi-project bridge data and trains industry-level AI models under privacy protection.

[0017] Preferably, the data acquisition layer contains a multi-source data space-time alignment middleware, the middleware includes a multi-source data access module, a space-time reference calibration module and a data fusion and registration module, the multi-source data access module adapts to TCP / IP and MQTT industrial protocols, and parses the space-time metadata of laser radar point cloud, stress data and unmanned aerial vehicle image;

[0018] The space-time reference calibration module solves the device pose through GNSS / SLAM fusion, realizes the conversion of the device coordinate system to the engineering global coordinate system, and realizes the multi-source device clock synchronization based on the formula P global = T S2G * T D2S * P device , wherein P global is a global coordinate, T S2G is a sensor-to-global coordinate system conversion matrix, T D2S is a device-to-sensor coordinate system conversion matrix, and P device is a device coordinate.

[0019] The data fusion and registration module uses ICP algorithm combined with plane constraint, i.e. to complete point cloud and image registration, and realizes the alignment of stress data and space data through timestamp interpolation, wherein p i is a point cloud coordinate, q i and r i are image plane coordinates, and ax+by+cz+d=0 is a plane equation.

[0020] Preferably, the space-time alignment middleware supports dynamic space-time reference switching, automatically switches the coordinate system according to the construction stage, and dynamically adjusts the data fusion weight through a multi-modal data weight self-adaptive algorithm to ensure the alignment accuracy in complex scenes.

[0021] Preferably, the AR headset of the client perception layer integrates binocular vision and IMU tight coupling algorithm, dynamically labels construction deviation through real-time superposition of BIM model, gesture / voice interaction control, wherein the binocular vision positioning is based on the formula Z is the target depth, f is the focal length, B is the baseline distance, and d is the parallax.

[0022] Preferably, the UAV group of the client perception layer carries a lightweight AI model to realize edge disease preliminary detection, only abnormal data is uploaded to the cloud, and the UWB cluster obstacle avoidance algorithm improves the obstacle avoidance response, and the obstacle avoidance algorithm formula is The distance D between the UAV and the obstacle is calculated, (x1, y1, z1) is the UAV coordinate, and (x2, y2, z2) is the obstacle coordinate.

[0023] Preferably, the blockchain storage technology of the ecological safety layer adopts a PBFT consensus algorithm.

[0024] Preferably, the construction management module of the logical processing layer introduces a disturbance factor model, dynamically adjusts the process every 5 minutes through a PPO algorithm, and the disturbance factor model formula is S = alpha * W + beta * E + gamma * M, wherein S is the construction state, W is the weather factor, E is the equipment factor, M is the personnel factor, and alpha, beta and gamma are weight coefficients.

[0025] The beneficial effects in the application are:

[0026] 1. Through the AR head-mounted display, UAV group and sensor network configured by the client perception layer, multi-source data acquisition in the whole bridge construction process is realized, and millimeter-level precision alignment of construction monitoring is achieved by using AR space anchoring and multi-sensor fusion technology. The multi-source data space-time alignment middleware of the data acquisition layer, by means of GNSS / SLAM fusion solution, PTP protocol clock synchronization and ICP algorithm combined with plane constraint registration mode, ensures the accurate unification of multi-modal data in space-time dimension. This high-precision data acquisition and alignment provides reliable data support for the construction process, enabling construction personnel to real-time and accurately grasp the construction status, timely discover and solve problems, avoid construction deviation and quality hidden dangers caused by data errors, and significantly improve the accuracy and reliability of construction monitoring.

[0027] 2. The business process engine constructed by setting the logical processing layer realizes the automation of construction deviation early warning and process compliance inspection, drives the closed-loop control of the bridge construction process, and the construction management module uses the PPO algorithm and the disturbance factor model to dynamically optimize the process according to the real-time construction progress, resource availability, weather conditions and other factors, and relies on the risk knowledge graph to realize real-time analysis and rapid early warning of construction risks, effectively avoiding resource waste and time delay in the construction process, improving the flexibility and adaptability of the construction plan, reducing the construction safety risk, making the construction management more intelligent and efficient, and ensuring the timely and high-quality completion of the project.

[0028] 3、By setting up intelligent analysis layer integrated finite element construction simulation, LSTM disease prediction and other models, the decision-making value of the bridge data can be deeply mined. The digital twin modeling platform realizes virtual pre-rehearsal of the construction process through hybrid precision modeling technology and ANSYS finite element kernel, discovers potential structural risks in advance and optimizes the construction scheme. The model cluster uses random forest algorithm and reinforcement learning algorithm, combines federal learning to aggregate multi-project data, realizes accurate prediction of construction progress, and these functions help the engineering team to fully simulate and plan before construction, dynamically adjust according to real-time data during construction, optimize resource scheduling, pre-rehearse extreme working conditions, improve construction safety and efficiency, reduce engineering changes and cost waste, and maximize engineering benefits.

[0029] 4、The zero trust architecture and collaborative platform built by setting up the ecological security layer realize cross-organization data sharing and full-process traceability through blockchain storage technology, dynamic permission control and other means, protect the bridge data security and multi-party cooperation, and the distributed storage adopts a hybrid architecture for efficient management of different types of data. The security architecture effectively resists data leakage and illegal access through device identity, space-time location, and biological feature triple verification and AES-256 encryption storage. The federal learning ecology aggregates multi-project data to train AI models under privacy protection, promoting industry technology progress. These measures not only ensure the safety and reliability of engineering data, but also break down information barriers, promote collaboration between all parties, improve the transparency and efficiency of engineering management, and provide a solid guarantee for the smooth implementation and long-term operation of the bridge project.

[0030] 5、Through the use of multi-modal data fusion prediction model by setting up the operation and maintenance detection module, combined with digital twin technology, the bridge disease can be accurately identified and the disease development trend can be predicted to realize predictive maintenance. When detecting the disease, the system automatically highlights the disease location on the digital twin model and generates a maintenance work order, greatly shortening the maintenance response time. This intelligent operation and maintenance management method significantly improves the disease identification accuracy and maintenance efficiency compared with traditional operation and maintenance methods, can timely handle potential bridge problems, avoid disease deterioration, effectively extend the service life of the bridge, reduce the whole life cycle maintenance cost, and ensure the long-term safe and stable operation of the bridge. BRIEF DESCRIPTION OF DRAWINGS

[0031] Fig. 1 A flowchart of a bridge information intelligent management system is proposed for the present application;

[0032] Fig. 2 An architecture diagram of a bridge information intelligent management system is proposed for the present application. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments.

[0034] With reference to Figs. 1-2 A bridge construction information intelligent management system, the management system comprising a client perception layer, a data collection layer, a logic processing layer, an intelligent analysis layer and an ecological security layer, the client perception layer being configured with an AR head-mounted display, an intelligent construction terminal, a drone group and a sensor network, for collecting bridge construction full-process multi-source data, realizing millimeter-level precision alignment of construction monitoring through AR space anchoring and multi-sensor fusion technology, covering beam hoisting and environment perception bridge construction scenes.

[0035] The data collection layer is deployed with a multi-protocol gateway and distributed storage, for cleaning and fusing bridge construction multi-modal data, unifying the construction coordinate system reference through a space-time alignment algorithm and a blockchain storage technology, correlating procedure timestamps, and ensuring traceability of bridge construction data throughout the process.

[0036] The logic processing layer constructs a business process engine, for connecting bridge construction design, construction and operation and maintenance businesses, automatically triggering construction deviation early warning and procedure compliance checking through BIM model verification and dynamic decision algorithm, and driving closed-loop control of bridge construction processes.

[0037] The intelligent analysis layer integrates simulation and prediction models, for mining bridge construction data decision value, optimizing bridge construction resource scheduling, and improving construction safety and efficiency through finite element construction simulation and LSTM disease prediction technology.

[0038] The ecological security layer builds a zero-trust architecture and a collaboration platform, for ensuring bridge construction data security and multi-party collaboration, realizing cross-organizational data sharing through a blockchain smart contract and dynamic permission control technology, and meeting bridge construction compliance audit requirements.

[0039] In the present embodiment, the logic processing layer comprises a design collaboration module, an operation and maintenance detection module, a construction management module and a collaborative management module, the design collaboration module dynamically updating a model through a parameterized BIM engine and realizing collision detection in combination with VR gesture interaction review, the operation and maintenance detection module improving disease identification accuracy through a multi-modal data fusion prediction model and combining with a digital twin body to real-time map a bridge state, the construction management module optimizing procedures through a PPO algorithm, adjusting resource priorities and relying on a risk knowledge graph to keep early warning response time at a level of hours, and the collaborative management module realizing multi-party data sharing through a dynamic data sandbox.

[0040] Specifically, in the design phase, the design coordination module receives parameters input by the designer, such as bridge span, beam size, material properties, etc. The parameterized BIM engine automatically generates a three-dimensional bridge model according to the formula, and reviews the model through VR gesture interaction. Design team members can enter the virtual environment and use gestures to rotate, scale, and section the model. Real-time design problems are marked, and the system automatically checks for collisions between bridge components and the surrounding environment through a collision detection algorithm. For example, the position conflict between the bridge pier foundation and the underground pipeline is avoided in advance to avoid design defects. Compared with traditional two-dimensional paper design, the design coordination module improves the design problem detection rate, reduces the number of design changes, and shortens the design cycle.

[0041] The operation and maintenance detection module continuously collects sensor data such as structural stress, displacement, and vibration during the bridge operation phase. Combined with unmanned aerial vehicle inspection images and AR head-mounted field detection data, the multi-modal data fusion prediction model CNN-LSTM network extracts and fuses different types of data features to accurately identify bridge cracks, corrosion, and other diseases. The digital twin body synchronously reflects the physical state of the bridge in real time. When a disease is detected, the system automatically highlights the disease location on the digital twin model and generates a maintenance work order, reducing the maintenance response time and delaying the service life of the bridge. The above multi-modal data fusion can simultaneously process two different types of multi-modal data, image data and time series data, breaking the limitations of a single data type and obtaining bridge state information from multiple dimensions. Moreover, due to the processing advantage of LSTM for time series data, the CNN-LSTM network can process dynamic data collected by sensors in real time, combined with image information, not only can identify existing diseases, but also can predict the development trend of diseases, providing strong support for preventive maintenance of bridges, such as predicting the expansion speed of cracks in advance to take timely measures;

[0042] CNN is a deep learning model designed specifically for handling data with grid structure. Its core components include convolutional layers, pooling layers, and fully connected layers. In processing unmanned aerial vehicle inspection images and AR head-mounted field detection images, the convolutional layer slides over the image with different convolutional kernels to extract local features such as crack shape, direction, and corrosion area texture. The pooling layer down-samples the feature map output by the convolutional layer to reduce data dimension and computational complexity while preserving key features. After stacking multiple convolutional-pooling layers, CNN can extract abstract and representative high-level features from the original image.

[0043] LSTM is a variant of recurrent neural networks, mainly used for processing sequence data, solving the problem of gradient disappearance and gradient explosion in traditional RNN, and can effectively capture the time dependence in long sequence. In operation and maintenance detection, bridge structure stress, displacement, vibration and other sensor data are typical time series data. LSTM can selectively remember or forget past information through cell state and gating mechanism, so as to learn the trend and rule of sensor data change over time, for example, to analyze the stress fluctuation pattern of bridge under different traffic load and weather conditions.

[0044] During construction, the construction management module uses PPO algorithm to dynamically optimize the sequence of construction processes according to real-time construction progress, resource availability, weather conditions and other factors. For example, when it detects rain, the algorithm automatically adjusts the outdoor work to indoor material processing and other non-weather affected processes. Relying on the risk knowledge graph, the system analyzes the construction risks in real time, such as the risk of slope instability during deep foundation pit excavation. Once the risk index exceeds the threshold, the system triggers an early warning within 1 hour and provides a risk response plan.

[0045] PPO algorithm makes real-time decisions based on real-time collected construction data such as progress, resource availability, weather conditions, etc. For example, through sensors or external weather information, PPO can automatically adjust the construction plan according to the detected rainfall prediction, such as adjusting the work that should be done outdoors to indoor or other weather-affected tasks. Through PPO algorithm, the system can find the optimal switching scheme between multiple processes, ensuring maximum resource utilization and minimizing construction delay. For example, when human or equipment resources are not available at a certain time, PPO can adjust the work order based on the current state to make the construction process as efficient and smooth as possible. The risk knowledge graph models various potential risks in the construction process and evaluates the risks based on historical data and expert experience. The system will evaluate and predict the risks that may occur during construction in real time according to the construction progress and environmental conditions. Combined with PPO algorithm, the risk of each key node in the construction process is dynamically calculated. If a certain risk index reaches a certain threshold, the system will automatically trigger an early warning within 1 hour, send a warning notification, and automatically generate a response plan based on historical risk data and knowledge graph. For example, if there is a risk of slope landslide in the deep foundation pit area, the system may suggest reinforcing the slope in time or suspending related work and transferring the construction team. The system can provide real-time visualization of construction progress, resource allocation, risk assessment and other data, which is convenient for project managers to monitor the construction status in real time and make immediate adjustments. Through dynamic updating of construction plans and risk warnings, construction managers can quickly identify problems and take appropriate measures.

[0046] Risk knowledge graph is an intelligent system that integrates, correlates and analyzes risk information using graph structure technology. It transforms scattered risk data into a semantic network of interconnected nodes, revealing complex relationships between risk factors and helping organizations more efficiently identify, assess and respond to various risks.

[0047] Resource availability risk refers to potential threats caused by disruptions or shortages in the supply of critical resources, which may hinder the achievement of business objectives. Risk knowledge graph can identify, assess and warn about such risks by structuring and correlating resource entities and their dependencies.

[0048] PPO algorithm is a reinforcement learning algorithm based on policy gradient, which improves the stability and efficiency of policy optimization by limiting the magnitude of policy updates. Its core idea is to use KL divergence to measure the difference between two policies and set a threshold to ensure that the difference between the new policy and the old policy does not exceed this threshold.

[0049] The collaborative management module provides a data access space separated by permissions for each party through a dynamic data sandbox, effectively improving work efficiency and data security. Design, construction and supervision units interact and collaborate within their respective permission ranges, ensuring information transparency and real-time feedback, avoiding information lag and misunderstandings. Through operation records that leave traces throughout the process, the data's traceability and responsibility traceability are enhanced, ensuring data security. In addition, the module ensures construction compliance, promptly identifies and resolves issues, reduces project risks and improves overall project management efficiency.

[0050] In this embodiment, the intelligent analysis layer includes a digital twin modeling platform, cloud computing and a model cluster. The digital twin modeling platform balances computational efficiency and accuracy through hybrid precision modeling technology, integrates ANSYS finite element kernel and supports virtual pre-performance of the construction process. Cloud computing deploys TensorRT inference engine and handles bridge displacement adjustment real-time control tasks through edge layer. The cloud executes large-scale finite element simulation and model training through Kubernetes elastic scheduling of GPU clusters. The model cluster progress prediction absorbs construction data in real time through random forest algorithm and reinforcement learning algorithm and aggregates multi-project data relying on federated learning.

[0051] Specifically, based on the BIM model, the hybrid precision modeling technology is used to simulate the connection between the pier and the beam body at key positions such as millimeter-level precision, ensuring high precision of structural analysis. Other non-critical positions are modeled with lower precision, greatly reducing the computational burden. After integrating the ANSYS finite element kernel, the digital twin platform can perform virtual pre-performance before construction, simulating the stress deformation of the bridge structure under different construction schemes.

[0052] By simulating the changes in the lifting point position during the lifting process and analyzing their impact on the stress of the beam body, the platform can optimize the construction plan based on these data and implement fine-grained control during actual construction. Combined with cloud computing and edge computing technologies, the platform can process real-time data from the construction site and perform large-scale finite element simulation and model training through Kubernetes' elastic scheduling of GPU clusters, further improving the accuracy of progress prediction during construction.

[0053] ANSYS Finite Element Kernel: ANSYS is a widely used engineering analysis and simulation software, particularly in the field of Finite Element Analysis (FEA). FEA is a numerical method used to simulate the behavior of physical systems (such as mechanics, thermodynamics, etc.) by dividing structures or objects into small elements to solve problems. In ANSYS, the finite element kernel is its core computing module, responsible for handling complex engineering simulation problems. It can simulate the mechanical properties of various materials, stress deformation, etc., providing support for design and optimization. In bridge construction simulation, the ANSYS finite element kernel is used to simulate the stress and deformation of the bridge structure under different construction schemes, helping designers optimize construction schemes.

[0054] Kubernetes Elastic Scheduling GPU Cluster: Kubernetes is an open-source container orchestration platform that automates the deployment, management, and scaling of containerized applications. Its elastic scheduling feature refers to Kubernetes dynamically adjusting the deployment and load distribution of containers based on the application's needs and resource availability. For GPU clusters, Kubernetes can effectively manage and schedule computing tasks on multiple GPU nodes, ensuring efficient resource usage and avoiding resource idleness or overload. In large-scale computing tasks such as finite element simulation and model training, Kubernetes' elastic scheduling can ensure optimal allocation of GPU resources, improving computing efficiency and saving costs.

[0055] TensorRT Inference Engine: TensorRT is a high-performance inference engine developed by NVIDIA, specifically designed to optimize and accelerate the inference process of deep learning models. Inference refers to the process of using a trained neural network model to make predictions on actual data. TensorRT uses techniques such as graph optimization, layer fusion, and precision adjustment to improve inference speed and performance. It supports multiple hardware platforms (especially NVIDIA's GPUs) and provides low-latency, high-throughput solutions for deep learning inference. In real-time control tasks during construction, the TensorRT inference engine can accelerate the inference process of machine learning models, quickly process real-time data and respond, such as adjusting the displacement of the bridge erecting machine or other construction equipment operations.

[0056] In this embodiment, the ecological security layer includes a distributed storage, a security architecture and a federated learning ecology. The distributed storage adopts a hybrid architecture, PostgreSQL stores structured work orders / logs, HDFS stores unstructured point cloud images, TimescaleDB stores time series sensor data, ensuring efficient management of multiple types of data. The security architecture verifies through device identity, space-time location and biological characteristics, dynamically adjusts data access permissions, and stores core data using AES-256 encryption. The federated learning ecology aggregates bridge data from multiple projects to train industry-level AI models under privacy protection.

[0057] Specifically, the hybrid architecture of distributed storage ensures efficient storage and management of different types of data. PostgreSQL, as a relational database, can handle structured work orders and log information, ensuring data reliability and consistency. HDFS, as a distributed file system, can store large amounts of unstructured data such as point cloud images, meeting the storage needs of massive data. TimescaleDB excels in handling time series data and is suitable for collected sensor data, providing fast data insertion and efficient query capabilities.

[0058] The security architecture enhances data security through multiple verification mechanisms. Device identity verification ensures that only authorized devices can access data. Space-time location verification monitors the geographic location of data sources, and biological feature verification further ensures the accuracy and security of data access. These measures work together to prevent unauthorized access and data leakage. Core data is stored using the AES-256 encryption algorithm to ensure that even if data is illegally obtained, the data itself cannot be interpreted.

[0059] The federated learning ecology aggregates data from multiple bridge projects to train models without revealing raw data, ensuring data privacy while also utilizing data from different projects to collaboratively train more efficient and accurate industry-level AI models. Through federated learning, model updates and optimizations do not require centralized data, but rather are calculated at each data source, with model parameters ultimately being aggregated, thereby implementing a decentralized learning approach that protects user privacy while enhancing overall model performance.

[0060] In this embodiment, the data acquisition layer includes a multi-source data space-time alignment middleware, which includes a multi-source data access module, a space-time reference calibration module, and a data fusion and registration module. The multi-source data access module adapts to TCP / IP and MQTT industrial protocols, parses the space-time metadata of laser radar point clouds, stress data, and unmanned aerial vehicle images. This module is responsible for adapting and parsing data transmitted by various industrial protocols, especially laser radar point clouds, stress data, and unmanned aerial vehicle images. Their space-time metadata needs to be extracted and standardized for uniform space-time alignment in subsequent processing.

[0061] The space-time reference calibration module calculates the device pose by fusing GNSS / SLAM, realizes the conversion from the device coordinate system to the global engineering coordinate system, and based on the formula P global = T S2G * T D2S * P device , where P global is the global coordinate, T S2G is the sensor-to-global coordinate system conversion matrix, T D2S is the device-to-sensor coordinate system conversion matrix, and P device is the device coordinate. Based on PTP protocol, it realizes multi-source device clock synchronization. This module calculates the device pose by fusing GNSS (Global Navigation Satellite System) and SLAM (Simultaneous Localization and Mapping) technology, and converts the device coordinate system to the global engineering coordinate system. Based on PTP (Precision Time Protocol), it realizes multi-source device clock synchronization, ensuring that all device-collected data can be accurately aligned in time.

[0062] The data fusion and registration module uses ICP algorithm combined with plane constraint, i.e. to complete point cloud and image registration, and through timestamp interpolation to realize stress data and space data alignment, where p i is the point cloud coordinate, q i and r i are image plane coordinates, and ax+by+cz+d=0 is the plane equation.

[0063] Specifically, the task of the data fusion and registration module is to align and register point clouds, images, and stress data. To achieve this goal, ICP (Iterative Closest Point) algorithm is used, which is a common point cloud registration method that aligns point clouds by minimizing the distance between them. Combined with plane constraint, it enables point clouds and image data to be accurately aligned, further improving data accuracy.

[0064] At the same time, based on the timestamp interpolation method, stress data and space data are space-time aligned to ensure that the position and stress state of the data at the same time can accurately reflect.

[0065] Through the cooperative work of the above modules, the data acquisition layer can realize efficient data access, time and space calibration, and fusion and registration of multi-source data, providing accurate time and space information for subsequent data analysis and processing.

[0066] In this embodiment, the spatio-temporal alignment middleware supports dynamic spatio-temporal reference switching, automatically switches coordinate systems according to construction stages, and adjusts data fusion weights dynamically through a multi-modal data weight adaptive algorithm to ensure the alignment accuracy in complex scenarios.

[0067] Specifically, according to different stages of the construction site, the spatio-temporal alignment middleware can automatically identify and switch different coordinate systems. In the early stage of construction, a local coordinate system may be used for data collection, while in the later stage of construction or the completion stage, a global coordinate system is used. This switching process is automatically triggered based on the construction progress, ensuring that construction data at different stages can be analyzed and fused in a unified coordinate system; different types of data have different effects on the final result in the alignment process. To address this difference, the system introduces an adaptive algorithm based on data quality and reliability. For example, in sunny weather, the quality of unmanned aerial vehicle image data is high, and the positioning error is ≤8 cm, so the weight is automatically set to 0.6; when the image is blurred due to rain, the positioning error increases to 15 cm, and the algorithm automatically reduces the image weight to 0.3, while increasing the weight of laser radar point cloud data to ensure the overall accuracy of data fusion.

[0068] In this embodiment, the AR headset of the client perception layer integrates binocular vision and IMU tight coupling algorithm, dynamically labels construction deviation through real-time superposition of BIM model, gesture / voice interaction control, etc. Z is the target depth, f is the focal length, B is the baseline distance, and d is the parallax.

[0069] Specifically, when the construction personnel wear AR headsets for work, the binocular vision system collects real-time images, calculates the depth information of the target object according to the triangulation principle formula, and obtains the attitude information of the headset through the IMU inertial measurement unit, realizing high-precision spatial positioning, with a virtual-real alignment accuracy of ≤2 mm. The construction personnel can superimpose the design information in the BIM model onto the real scene in real time through gestures or voice commands.

[0070] In this embodiment, the UAV group of the client perception layer carries a lightweight AI model to realize edge disease preliminary detection, and only uploads abnormal data to the cloud and uses the UWB cluster obstacle avoidance algorithm to improve the obstacle avoidance response. The distance D between the UAV and the obstacle is calculated, and (x1, y1, z1) is the UAV coordinate, and (x2, y2, z2) is the obstacle coordinate.

[0071] Specifically, in the bridge inspection task, each UAV carries a lightweight AI model YOLOv8-Tiny, which analyzes the collected image data in real time at the edge, quickly identifies bridge surface cracks, holes and other diseases, and for areas where no abnormalities are detected, the UAV only saves key frame images and does not upload data to the cloud; when an abnormal disease is detected, the high-definition image and disease analysis result are uploaded to the cloud for further accurate diagnosis. At the same time, the UAV group uses the UWB cluster obstacle avoidance algorithm to calculate the distance D between the UAV and the obstacle in real time, and when the distance is less than the safety threshold, the obstacle avoidance action is triggered within 0.1s to adjust the flight attitude and avoid collision.

[0072] In this embodiment, the blockchain storage technology of the ecological security layer adopts the PBFT consensus algorithm.

[0073] Specifically, in the process of storing bridge construction data, the blockchain storage technology adopts the PBFT (practical Byzantine fault tolerance) consensus algorithm, and each construction unit, supervision unit and owner unit node participates in data verification and block generation. When new construction data such as concealed engineering acceptance records and material detection reports need to be stored, the initiating node broadcasts the data to the network, and other nodes receive the data and verify it. The nodes that pass the verification complete the block generation and consensus process within 2 seconds according to the PBFT algorithm rules. Once the data is stored on the chain, it is encrypted and stored using the SHA-256 hashing algorithm. Data tampering will change the hash value, and the system can detect data tampering. This technology ensures the integrity and credibility of engineering data, and can quickly provide tamper-proof evidence in engineering dispute resolution.

[0074] In this embodiment, the construction management module of the logical processing layer introduces a disturbance factor model, which dynamically adjusts the process every 5 minutes through the PPO algorithm. The disturbance factor model formula is S=α*W+β*E+γ*M, where S is the construction state, W is the weather factor, E is the equipment factor, M is the personnel factor, and α, β, and γ are weight coefficients.

[0075] Specifically, the construction management module collects real-time weather data such as rainfall and wind speed, equipment operation data such as bridge girder erection machine failure frequency and tower crane working status, and personnel data such as the attendance rate of personnel of various types. According to the disturbance factor model formula S=α*W+β*E+γ*M, the current construction state comprehensive score S is calculated. When rainfall causes the value to increase and bridge girder erection machine failure causes the value E to rise, the S value exceeds the normal range, and the system determines that the construction state is abnormal. At this time, the PPO algorithm re-plans the process every 5 minutes according to the current construction state, and adjusts outdoor work affected by the weather to indoor work.

[0076] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, makes equivalent replacement or change within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. An intelligent management system for bridge construction information, characterized in that: The management system includes a client perception layer, a data acquisition layer, a logic processing layer, an intelligent analysis layer, and an ecological security layer. The client perception layer is equipped with an AR head-mounted display, an intelligent construction terminal, a drone swarm, and a sensor network to collect multi-source data throughout the entire bridge erection process. Through AR spatial anchoring and multi-sensor fusion technology, it achieves millimeter-level precision alignment for construction monitoring, covering beam hoisting and environmental perception bridge erection scenarios. The data acquisition layer deploys a multi-protocol gateway and distributed storage to clean and fuse multimodal bridge construction data. Through spatiotemporal alignment algorithms and blockchain storage technology, it unifies the construction coordinate system benchmark and associates process timestamps to ensure that the entire bridge construction data process is traceable. The logic processing layer constructs a business process engine to connect bridge design, construction, and operation and maintenance. Through BIM model verification and dynamic decision-making algorithms, it automatically triggers construction deviation warnings and process compliance checks, driving closed-loop control of the bridge construction process. The intelligent analysis layer integrates simulation and prediction models to extract the decision-making value of bridge construction data. Through finite element construction simulation and LSTM defect prediction technology, it optimizes bridge construction resource scheduling, simulates extreme working conditions, and improves construction safety and efficiency. The aforementioned ecological security layer establishes a zero-trust architecture and collaborative platform to ensure the security of bridging data and multi-party collaboration. Through blockchain smart contracts and dynamic permission control technology, it enables cross-organizational hierarchical data sharing and meets the bridging compliance audit requirements.

2. The intelligent management system for bridge construction information according to claim 1, characterized in that: The logical processing layer includes a design collaboration module, an operation and maintenance monitoring module, a construction management module, and a collaborative management module. The design collaboration module dynamically updates the model through a parametric BIM engine and combines VR gesture interaction for review to achieve collision detection. The operation and maintenance monitoring module improves the accuracy of defect identification through a multimodal data fusion prediction model and combines the construction of a digital twin to map the bridge status in real time. The construction management module optimizes work processes and adjusts resource priorities through the PPO algorithm and relies on a risk knowledge graph to keep the early warning response time at the hour level. The collaborative management module uses a dynamic data sandbox to achieve multi-party data sharing.

3. The intelligent management system for bridge construction information according to claim 1, characterized in that: The intelligent analysis layer includes a digital twin modeling platform, cloud computing, and a model cluster. The digital twin modeling platform balances computational efficiency and accuracy through hybrid precision modeling technology, integrates the ANSYS finite element kernel, and supports virtual pre-simulation of the construction process. The cloud computing deploys the TensorRT inference engine at the edge layer and handles real-time control tasks for bridge erecting machine displacement adjustment. The cloud uses Kubernetes to elastically schedule GPU clusters to perform large-scale finite element simulations and model training. The model cluster's progress prediction uses random forest and reinforcement learning algorithms to absorb construction data in real time and aggregates data from multiple projects through federated learning.

4. The intelligent management system for bridge construction information according to claim 1, characterized in that: The ecological security layer includes distributed storage, a security architecture, and a federated learning ecosystem. The distributed storage adopts a hybrid architecture: PostgreSQL stores structured data products / logs, HDFS stores unstructured point cloud images, and TimescaleDB stores time-series sensor data, ensuring efficient management of multiple data types. The security architecture dynamically adjusts data access permissions through triple verification of device identity, spatiotemporal location, and biometrics. Core data is stored using AES-256 encryption. The federated learning ecosystem aggregates bridge data from multiple projects to train industry-level AI models under privacy protection.

5. The intelligent management system for bridge construction information according to claim 1, characterized in that: The data acquisition layer includes a multi-source data spatiotemporal alignment middleware. The middleware includes a multi-source data access module, a spatiotemporal reference calibration module, and a data fusion and registration module. The multi-source data access module is compatible with TCP / IP and MQTT industrial protocols and can parse spatiotemporal metadata of lidar point clouds, stress data, and UAV imagery. The spatiotemporal reference calibration module calculates the device pose through GNSS / SLAM fusion, realizing the transformation from the device coordinate system to the engineering global coordinate system, based on formula P. global =T S2G *T D2S *P device , where P global For global coordinates, T S2G T is the transformation matrix from the sensor to the global coordinate system. D2S P is the transformation matrix from device to sensor coordinate system. device Use device coordinates and implement multi-source device clock synchronization based on the PTP protocol; The data fusion and registration module employs the ICP algorithm combined with planar constraints, namely... Point cloud and image registration were completed, and stress data and spatial data were aligned through timestamp interpolation, where p i Let q be the point cloud coordinates. i r i Let ax + by + cz + d = 0 be the coordinates of the image plane, and let ax + by + cz + d = 0 be the equation of the plane.

6. The intelligent management system for bridge construction information according to claim 5, characterized in that: The spatiotemporal alignment middleware supports dynamic spatiotemporal reference switching, automatically switching coordinate systems according to the construction stage, and utilizes a multimodal data weight adaptive algorithm. Dynamically adjust data fusion weights to ensure alignment accuracy in complex scenarios.

7. The intelligent management system for bridge construction information according to claim 1, characterized in that: The AR headset in the client-side perception layer integrates binocular vision and IMU tightly coupled algorithms, dynamically marking construction deviations through real-time overlay of BIM models and gesture / voice interaction control; wherein the binocular vision positioning is based on the triangulation principle formula. Z represents the target depth, f represents the focal length, B represents the baseline distance, and d represents the parallax.

8. The intelligent management system for bridge construction information according to claim 1, characterized in that: The client-side perception layer, equipped with a lightweight AI model, performs initial edge-end defect detection, uploading only abnormal data to the cloud, and uses a UWB cluster obstacle avoidance algorithm to improve obstacle avoidance response. The obstacle avoidance algorithm formula is described below. Calculate the distance D between the drone and the obstacle, where (x1, y1, z1) are the coordinates of the drone and (x2, y2, z2) are the coordinates of the obstacle.

9. The intelligent management system for bridge construction information according to claim 1, characterized in that: The blockchain-based evidence storage technology of the ecological security layer adopts the PBFT consensus algorithm.

10. The intelligent management system for bridge construction information according to claim 1, characterized in that: The construction management module of the logic processing layer introduces a disturbance factor model, which dynamically adjusts the work process every 5 minutes using the PPO algorithm. The formula for the disturbance factor model is S=α*W+β*E+γ*M, where S is the construction status, W is the weather factor, E is the equipment factor, M is the personnel factor, and α, β, and γ are weighting coefficients.

Images