A BIM-based optimized management system for hoisting construction supervision

Abstract

This invention discloses a BIM-based hoisting construction supervision and optimization management system, comprising: a perception layer: real-time acquisition of multi-source heterogeneous data; a decision layer: the bottom layer uses an incremental RRT# algorithm to generate candidate paths that satisfy the kinematic constraints of the crane, and the upper layer embeds a distributed Q-learning framework, where each crane acts as an independent intelligent agent, collaboratively learning through a shared experience pool, dynamically optimizing path planning strategies based on environmental perception data and construction progress requirements, and simultaneously developing a fuzzy sliding mode control algorithm combined with an LSTM-Transformer model of crane trolley and hook travel speeds and hook lifting speeds, as well as a wire rope disturbance prediction model to calculate crane motion compensation parameters, and generating optimal control commands in advance based on predicted crane movement and lifting data; an execution layer: the commands generated by the decision layer are issued to the construction equipment; and an optimization layer: a BIM model is constructed, and the equipment execution results and structural safety monitoring data are fed back to the decision layer.

Description

Technical Field

[0001] This invention relates to the field of safety technology for the hoisting of large components in pumped storage hydropower stations, and in particular to a BIM-based optimized management system for hoisting construction supervision. Background Technology

[0002] Pumped storage power station buildings have complex multi-layered spatial structures. The harsh construction environment for hoisting large components such as stators, rotors, runners, top covers, inlet valves, and GCBs (generator outlet switches) makes the hoisting of large components the most risky part of the process. The bridge crane (referred to as the bridge crane) in the main powerhouse is the core lifting equipment for large component hoisting. It consists of a bridge frame (also known as the trolley), hoisting mechanism, trolley, trolley traveling mechanism, operator's cab, trolley power supply device (auxiliary conductor rail), and main power supply device (main conductor rail). The trolley travels longitudinally along tracks laid on elevated structures on both sides, while the trolley travels laterally along tracks laid on the bridge frame. Traditional supervision models rely on manual experience and offline planning, making it difficult to handle complex scenarios involving multiple variables such as crane movement, lifting speed and wire rope disturbance, equipment coordination, and structural deformation. For example, hook swaying caused by the movement, lifting, or stopping of the crane's trolley or trolley can lead to collisions. Path conflicts between the main plant's bridge crane and other cranes operating on the same or different levels can reduce construction efficiency. Furthermore, the uncertainties of multi-faceted operations during initial installation or unit maintenance of a power station can pose safety hazards. The maturity of Building Information Modeling (BIM) technology provides a digital foundation for construction supervision. Its 3D visualization, data integration, and simulation analysis capabilities can precisely compensate for the shortcomings of traditional supervision in terms of real-time performance, accuracy, and predictability. By deeply integrating BIM with technologies such as the Internet of Things, artificial intelligence, and control theory, an intelligent supervision system covering the entire chain of "perception-decision-execution-optimization" can be constructed. This enables core functions such as dynamic path planning, real-time compensation for crane movement speed, lifting speed and wire rope disturbance, and structural safety early warning in hydropower station construction, especially pumped storage power stations. Ultimately, this significantly improves the safety margin and construction efficiency of lifting operations.

[0003] In existing technologies, the improved RRT* algorithm is tightly integrated with the BIM model. The BIM model provides rich building information, and the improved RRT* algorithm can more accurately plan paths that meet actual construction needs by calling upon this information. However, although the improved RRT* algorithm considers equipment kinematic constraints, the environment changes very rapidly in the civil construction and electromechanical equipment installation of pumped storage power stations (such as multiple work surfaces, temporary obstacles, etc.), and the real-time response capability of the algorithm may be insufficient. This may cause the construction equipment to continue to move along the original path, which may deviate from the actual working conditions and increase the risk of collision. Therefore, a BIM-based hoisting construction supervision and optimization management system is proposed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a BIM-based optimized management system for hoisting construction supervision.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A BIM-based hoisting construction supervision and optimization management system includes:

[0007] Perception Layer: Deploy 5G+ edge computing nodes to collect multi-source heterogeneous data from millimeter-wave radar, UWB+ lidar, micro lidar weather stations, and fiber optic grating sensor networks in real time. Use Kalman filter-particle filter fusion algorithm to construct a dynamic obstacle spatiotemporal map and align the spatiotemporal data of multiple sensors. Simultaneously, embed fiber optic grating sensor networks and deploy digital twin mirror interfaces to complete real-time monitoring of strain, tilt, and steel structure stress of key crane components and synchronize the BIM model with the physical entity status.

[0008] Decision layer: The bottom layer uses the incremental RRT# algorithm, combined with a non-uniform sampling strategy and intelligent pruning mechanism, to generate candidate paths that meet the kinematic constraints of the crane. The upper layer embeds a distributed Q-learning framework, in which the crane acts as an independent intelligent agent and learns collaboratively through a shared experience pool. Based on environmental perception data and construction progress requirements, the path planning strategy is dynamically optimized. At the same time, a fuzzy sliding mode control algorithm is developed and combined with the LSTM-Transformer crane trolley and crane travel speed, hook lifting speed and wire rope disturbance prediction model to calculate the crane motion compensation parameters. Model predictive control is introduced, and the optimal control command is generated in advance based on the predicted travel speed, lifting speed and wire rope disturbance data.

[0009] Execution layer: The path planning instructions and wire rope disturbance compensation control instructions generated by the decision layer are sent to the construction equipment through the OPC UAover TSN protocol;

[0010] Optimization layer: Construct a BIM model, integrate geometric information, construction progress, structural response and equipment status, and feed back the equipment execution results and structural safety monitoring data to the decision-making layer.

[0011] The above technical solution further includes:

[0012] Furthermore, the multi-source heterogeneous data collected by the perception layer includes environmental perception data, structural perception data, equipment status data, and personnel and safety data. The environmental perception data includes millimeter-wave radar data, UWB+ lidar data, and micro lidar weather station data. The structural perception data includes fiber optic grating sensor network data and digital twin mirror interface data. The equipment status data includes crane operation status data and elevator operation status data. The personnel and safety data includes personnel positioning data and safety monitoring video data.

[0013] Furthermore, the perception layer employs a Kalman filter-particle filter fusion algorithm to perform the following specific steps for constructing a dynamic obstacle spatiotemporal map and aligning multi-sensor data spatiotemporally:

[0014] Data preprocessing: Through data cleaning and data format standardization, outliers are detected and removed using thresholding or statistical methods, and a unified data structure is defined to perform preliminary processing on the raw sensor data, remove noise and outliers, and convert data from different sensors into a unified format;

[0015] Kalman filtering: It recursively estimates the state of a linear Gaussian system by using the state transition matrix and the observation matrix through two steps: prediction and update.

[0016] Particle filtering: It represents the state distribution of a system through a set of random samples and approximates the true posterior distribution through resampling and weight updates, thus completing the state estimation of a nonlinear non-Gaussian system.

[0017] Fusion: The state estimation of the linear part is performed using Kalman filtering, and the state estimation of the nonlinear part is performed using particle filtering. The results of the two filters are then fused using a weighted average or an interactive multi-model algorithm.

[0018] Furthermore, the specific steps for the perception layer to perform digital twin mirror synchronization are as follows:

[0019] Sensor deployment: Deploy fiber Bragg grating sensor networks on critical components of the crane;

[0020] Integration of the digital twin mirror interface with the BIM model: Assign a unique identifier to each component of the BIM model, and also assign a unique identifier to each sensor in the sensor network. By establishing a mapping relationship between the identifiers, the sensor data is associated with the BIM model components. When the sensor data changes, the data is synchronized to the BIM model through the interface, driving the update of the information of the model components.

[0021] Furthermore, the underlying layer utilizes the incremental RRT# algorithm, combined with a non-uniform sampling strategy and an intelligent pruning mechanism, to generate candidate paths that satisfy the kinematic constraints of the crane, including the following steps;

[0022] Initialization: Set the starting point and target point in the BIM model;

[0023] Non-uniform sampling: Based on the risk heatmap, sampling is performed in key areas. The sampling probability P(x) is calculated based on the risk value R(x) at location x and the obstacle density D(x), expressed by the formula P(x)=(R(x)·D(x)) / (∫ Ω R(x′)·D(x′)dx′), where Ω is the sampling space, and the integral term is used to normalize the probability;

[0024] Path expansion: Using the incremental RRT# algorithm, the path is randomly expanded from the starting point, and the kinematic constraints of the crane are considered in each expansion;

[0025] Intelligent pruning: Removes redundant paths and retains paths with lower cost and that satisfy kinematic constraints;

[0026] Path optimization: Smoothing the preserved paths;

[0027] Output candidate paths: Output candidate paths that satisfy all constraints.

[0028] Furthermore, the upper layer embeds a distributed Q-learning framework, where each crane acts as an independent intelligent agent. Through collaborative learning via a shared experience pool, it dynamically optimizes path planning strategies based on environmental perception data and construction progress requirements, including the following steps:

[0029] Initialization: The agent obtains the current state information from the BIM model and initializes the Q-value table;

[0030] Experience collection: Each agent collects experience during the interaction process and stores it in a shared experience pool;

[0031] Collaborative learning: The agent randomly selects a batch of experiences from the experience pool for learning, and updates the Q-value table using the Q-learning algorithm based on the selected experiences;

[0032] Path planning: Based on the learned Q-value table, the agent selects the optimal action for path planning;

[0033] Dynamic adjustment: During the hoisting process, based on environmental perception data and construction progress requirements (such as adjustments to the floor construction progress), the agent selects the optimal action, executes the selected action, updates its own position, and receives reward signals from the environment based on the results of the action.

[0034] Furthermore, the developed fuzzy sliding mode control algorithm combines the LSTM-Transformer crane trolley and crane travel speeds, hook lifting speeds, and wire rope disturbance prediction models to calculate crane motion compensation parameters. Model predictive control is then introduced, and based on the predicted travel speeds, lifting speeds, and wire rope disturbance data, optimal control commands are generated in advance. This includes the following steps:

[0035] Fuzzy sliding mode control algorithm: A sliding surface function s(t) is defined to represent the deviation between the system state and the desired state. The sliding surface function is expressed as... Where e(t) is the position or angle error, Here, c is the error rate of change, and c is the design parameter. A control law u(t) is designed to make the system state reach the sliding surface in a finite time and move along the sliding surface to the equilibrium point. The control law is expressed as u(t) = u eq (t)+u sw (t), where u eq (t) is the equivalent control term, u sw (t) is the switching control term, and fuzzy logic is used to adjust the gain of the switching control term to adapt to different operating conditions;

[0036] Construct an LSTM-Transformer model for predicting the travel speed of the crane trolley and crane body, the lifting speed of the hook, and the wire rope disturbance: The LSTM layer processes time-series data to capture short-term changes in travel speed, lifting speed, and wire rope disturbance; the Transformer encoder processes spatial data to capture the spatial correlation of the large components when the travel speed, lifting speed, and wire rope disturbance occur at different locations; the model is trained using historical travel speed, lifting speed, and wire rope disturbance data, and the loss function is optimized to enable the model to predict the lifting speed and wire rope disturbance data for a future period of time.

[0037] Integrated Model Predictive Control: Based on the LSTM-Transformer crane trolley and crane travel speed, hook lifting speed, and wire rope disturbance prediction model, rolling optimization, and feedback correction, the system takes the current lifting speed and the position data of the large component as input and outputs the predicted values ​​of future travel speed, lifting speed, and wire rope disturbance. Within each control cycle, it solves the finite-time optimal control problem and uses an optimization algorithm to solve the optimal control problem, obtaining the optimal control sequence. The first element of the optimal control sequence is applied to the crane system, thereby generating the optimal control command in advance.

[0038] Furthermore, the step of sending the data to the construction equipment via the OPC UA over TSN protocol includes the following steps:

[0039] Instruction Encoding and Encapsulation: Encode the path planning results and compensation parameters into an OPC UA data structure;

[0040] OPC UA over TSN network transmission: Construct a construction equipment information model based on OPC UA, define the nodes and methods of equipment, sensors, and actuators, configure the time synchronization and traffic scheduling parameters of the TSN network, and send the encoded instructions to the construction equipment through the OPC UA over TSN protocol;

[0041] Equipment-side command decoding and execution: An OPC UA client is set up on the construction equipment to receive and decode commands from the decision-making level. The OPC UA client parses the command parameters in the OPC UA data structure according to the specific model and configuration of the equipment.

[0042] The present invention has the following beneficial effects:

[0043] In this invention, the bottom layer of the decision layer generates candidate paths that satisfy the kinematic constraints of the crane through a non-uniform sampling strategy and an intelligent pruning mechanism, thereby improving path search efficiency. The upper layer of the decision layer embeds a distributed Q-learning framework, and the crane, as an independent intelligent agent, performs collaborative learning through a shared experience pool. Based on environmental perception data and construction progress requirements, it dynamically optimizes the path planning strategy and can quickly converge to the optimal path when the environment changes abruptly. At the same time, a fuzzy sliding mode control algorithm is developed and combined with an LSTM-Transformer crane trolley and crane travel speed, hook lifting speed and wire rope disturbance prediction model to calculate the crane motion compensation parameters. Based on the predicted travel speed, lifting speed and wire rope disturbance data, the optimal control command is generated in advance to shorten the equipment response delay. Attached Figure Description

[0044] Figure 1 This is a system block diagram of a BIM-based hoisting construction supervision and optimization management system proposed in this invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 As shown, this invention is a BIM-based hoisting construction supervision and optimization management system, comprising:

[0047] Perception Layer: Deploys 5G+ edge computing nodes to collect multi-source heterogeneous data in real time from millimeter-wave radar (300m detection range, 50Hz refresh rate), UWB+ lidar (3D positioning accuracy 2cm), micro lidar weather station (10Hz sampling frequency), and fiber optic grating sensor network (strain resolution 1με, tilt measurement accuracy 0.01°). Employs Kalman filter-particle filter fusion algorithm to construct a dynamic obstacle spatiotemporal map and align the multi-sensor data in the spatiotemporal space. Simultaneously, embeds the fiber optic grating sensor network and deploys a digital twin mirror interface to complete real-time monitoring of strain, tilt, and steel structure stress of key crane components, as well as synchronization of the BIM model and physical entity status.

[0048] Decision layer: The bottom layer uses the incremental RRT# algorithm, combined with a non-uniform sampling strategy and intelligent pruning mechanism, to generate candidate paths that meet the kinematic constraints of the crane. The upper layer embeds a distributed Q-learning framework, where the crane acts as an independent intelligent agent and learns collaboratively through a shared experience pool. Based on environmental perception data and construction progress requirements, it dynamically optimizes the path planning strategy. At the same time, a fuzzy sliding mode control algorithm is developed and combined with the LSTM-Transformer crane trolley and trolley travel speed, hook lifting speed and wire rope disturbance prediction model to calculate the crane motion compensation parameters. Model predictive control (MPC) is introduced, and based on the predicted travel speed, lifting speed and wire rope disturbance data, the optimal control command is generated in advance. A physical constraint neural network is constructed, and the CFD simulation results are used as supervision signals to train the 3D-CNN+Graph Transformer model. The model parameters are dynamically updated based on real-time monitoring data.

[0049] Execution layer: The path planning instructions and lifting speed and wire rope disturbance compensation control instructions generated by the decision layer are sent to construction equipment such as cranes and elevators through the OPC UA over TSN protocol. A safety redundancy mechanism with dual PLC hot backup is deployed. When the main controller fails, the backup controller performs fault switching.

[0050] Optimization layer: A BIM model was constructed, integrating geometric information, construction progress, structural response and equipment status. Extreme working conditions were simulated through a digital twin testbed, and the equipment execution results and structural safety monitoring data were fed back to the decision-making layer to update the environmental perception model, path planning algorithm, movement and lifting speed and wire rope disturbance compensation control strategy, etc.

[0051] In one embodiment, the multi-source heterogeneous data collected by the perception layer includes environmental perception data, structural perception data, equipment status data, and personnel and safety data. The environmental perception data includes millimeter-wave radar data (using deployed millimeter-wave radar with a detection range of up to 300m and an update frequency of up to 50Hz to obtain real-time dynamic information such as distance, speed, and direction of obstacles around the construction site), UWB+LiDAR data (combining ultra-wideband (UWB) positioning technology and the 3D scanning capability of LiDAR to obtain high-precision 3D positioning data), and micro LiDAR weather station data (micro LiDAR weather stations deployed at the construction site monitor meteorological parameters such as wind speed, wind direction, temperature, and humidity in real time). The structural perception data... This includes fiber Bragg grating sensor network data (fiber Bragg grating sensor networks embedded in key crane components monitor the strain, tilt, and stress of the steel structure in real time) and digital twin mirror interface data (through the digital twin mirror interface, the status data of the BIM model and physical entities (such as cranes, elevators, etc.) are synchronized in real time, including geometric information, construction progress, equipment status, etc.). The equipment status data includes crane operating status data (including crane movement speed, lifting speed, luffing acceleration, lifting height, load weight, and other operating parameters) and elevator operating status data (elevator running speed, load capacity, position, etc.). The personnel and safety data includes personnel positioning data and safety monitoring video data.

[0052] In one embodiment, the perception layer employs a Kalman filter-particle filter fusion algorithm to perform the following specific steps for constructing a dynamic obstacle spatiotemporal map and aligning multi-sensor data spatiotemporally:

[0053] Data preprocessing: Through data cleaning and data format standardization, outliers are detected and removed using thresholding or statistical methods, and a unified data structure is defined to perform preliminary processing on the raw sensor data, remove noise and outliers, and convert data from different sensors into a unified format;

[0054] Kalman filtering: It recursively estimates the state of a linear Gaussian system using the state transition matrix and observation matrix through two steps: prediction and update. Prediction: State prediction uses the state transition matrix F and the optimal estimate from the previous time step. To predict the state at the current moment Covariance prediction: using the state transition matrix F and the covariance P from the previous time step. k-1 To predict the covariance P at the current moment k :P k =FP k-1 F T +Q, where Q is the covariance matrix of the process noise; Update: Calculate the Kalman gain: using the predicted covariance Pk The Kalman gain K is calculated using the covariance R of the observation noise. k :K k =P k H T HP k H T +R) -1 Where H is the observation matrix; State update: using Kalman gain K k Predicted value z k and predicted state To update the optimal state estimate at the current time step Covariance Update: Using Kalman Gain K k And the predicted covariance P k To update the covariance Pk at the current time step: Pk = (IK) k H)P k Where I is the identity matrix; Iteration: the optimal state estimate at the current time step is... The sum and covariance Pk are used as the initial state estimate and covariance for the next time step, and the prediction and update are repeated.

[0055] Particle filtering: It represents the state distribution of a system by a set of random samples (particles) and approximates the true posterior distribution by resampling and weight updates, thus completing the state estimation of a nonlinear non-Gaussian system.

[0056] Fusion: Kalman filtering is used for state estimation of the linear part, and particle filtering is used for state estimation of the nonlinear part. The results of the two filters are then fused using a weighted average or an interactive multi-model algorithm.

[0057] In one embodiment, the specific steps for the perception layer to perform digital twin mirror synchronization are as follows:

[0058] Sensor deployment: Deploy fiber Bragg grating sensor networks on key components of the crane, such as the main beam, trolley beam, and winch drum;

[0059] Integration of the digital twin mirror interface with the BIM model: Assign a unique identifier to each component of the BIM model, and also assign a unique identifier to each sensor in the sensor network. By establishing a mapping relationship between the identifiers, the sensor data is associated with the BIM model components. When the sensor data changes, the data is synchronized to the BIM model through the interface, driving the update of information such as the geometry, position, and attributes of the model components.

[0060] In one embodiment, the underlying layer utilizes the incremental RRT# algorithm, combined with a non-uniform sampling strategy and an intelligent pruning mechanism, to generate candidate paths that satisfy the kinematic constraints of the crane, including the following steps;

[0061] Initialization: Set the starting point (ground) and the target point (50th floor) in the BIM model;

[0062] Non-uniform sampling: Based on a risk heatmap (generated from historical data, expert knowledge, or real-time monitoring data, reflecting the risk level of different locations), sampling is conducted in key areas. The sampling probability P(x) is calculated based on the risk value R(x) at location x and the obstacle density D(x), expressed by the formula P(x) = (R(x)·D(x)) / (∫ Ω R(x′)·D(x′)dx′), where Ω is the sampling space, and the integral term is used to normalize the probability;

[0063] Path expansion: Using the incremental RRT# algorithm, the path is randomly expanded from the starting point, and the kinematic constraints of the crane are considered in each expansion;

[0064] Intelligent pruning: Removes redundant paths and retains paths with lower cost and that satisfy kinematic constraints;

[0065] Path optimization: Smooth the retained path to reduce abrupt angle changes and acceleration fluctuations;

[0066] Output candidate paths: Output candidate paths that satisfy all constraints.

[0067] In one embodiment, the upper layer embeds a distributed Q-learning framework, where the crane acts as an independent intelligent agent. Through collaborative learning via a shared experience pool, it dynamically optimizes its path planning strategy based on environmental perception data and construction progress requirements, including the following steps:

[0068] Initialization: The agent obtains the current state information from the BIM model and initializes the Q-value table;

[0069] Experience collection: Each agent collects experience during the interaction process and stores it in a shared experience pool;

[0070] Collaborative learning: The agent randomly selects a batch of experiences from the experience pool for learning, and updates the Q-value table using the Q-learning algorithm based on the selected experiences;

[0071] Path planning: Based on the learned Q-value table, the agent selects the optimal action for path planning to avoid collisions and conflicts;

[0072] Dynamic adjustment: During the hoisting of large components of the unit, based on environmental perception data (such as wind speed changes) and construction progress requirements (such as the construction progress of the generator floor or the adjustment of the installation progress of a large component of a unit), the agent selects the optimal action, executes the selected action, updates its own position, and receives reward signals from the environment based on the results of the action.

[0073] In one embodiment, the developed fuzzy sliding mode control algorithm combines the LSTM-Transformer crane trolley and crane travel speeds, hook lifting speeds, and wire rope disturbance prediction models to calculate crane motion compensation parameters, introduces model predictive control (MPC), and generates optimal control commands in advance based on predicted travel speeds, lifting speeds, and wire rope disturbance data, including the following steps:

[0074] Fuzzy sliding mode control algorithm: A sliding surface function s(t) is defined to represent the deviation between the system state and the desired state. The sliding surface function is expressed as... Where e(t) is the position or angle error, Here, c is the error rate of change, and c is the design parameter. A control law u(t) is designed to make the system state reach the sliding surface in a finite time and move along the sliding surface to the equilibrium point. The control law is expressed as u(t) = u eq (t)+u sw (t), where u eq (t) is the equivalent control term, u sw (t) is the switching control term. At the same time, fuzzy logic is used to adjust the gain of the switching control term to adapt to different working conditions, thereby effectively handling the uncertainty and nonlinearity in the crane system.

[0075] Construct an LSTM-Transformer model for predicting the travel speed of the crane trolley and crane hook lifting speed and wire rope disturbance: The LSTM layer processes time-series data to capture short-term changes in wind speed and direction, while the Transformer encoder processes spatial data to capture the spatial correlation between lifting speed and wire rope disturbance at different locations. The model is trained using historical travel speed, lifting speed and wire rope disturbance data, and the loss function is optimized to enable the model to predict wire rope disturbance data for a future period.

[0076] Integrated Model Predictive Control (MPC): Based on the LSTM-Transformer crane trolley and crane travel speed, hook lifting speed, and wire rope disturbance prediction model, rolling optimization, and feedback correction, the system takes the current travel speed and lifting speed data as input and outputs the predicted value of the future wire rope disturbance. In each control cycle, it solves the finite-time optimal control problem using optimization algorithms (such as gradient descent and genetic algorithms) to obtain the optimal control sequence. The first element of the optimal control sequence is applied to the crane system, and feedback correction ensures the accuracy of the prediction model, thereby generating the optimal control command in advance.

[0077] In one embodiment, the step of sending data to construction equipment such as cranes and construction elevators via the OPC UA over TSN protocol includes the following steps:

[0078] Instruction Encoding and Encapsulation: Encode the path planning results and compensation parameters into an OPC UA data structure;

[0079] OPC UA over TSN network transmission: Construct a construction equipment information model based on OPC UA, define the nodes and methods of equipment, sensors, and actuators, configure the time synchronization and traffic scheduling parameters of the TSN network, and send the encoded instructions to the construction equipment through the OPC UA over TSN protocol;

[0080] Equipment-side command decoding and execution: An OPC UA client is set up on the construction equipment to receive and decode commands from the decision-making level. The OPC UA client parses the command parameters in the OPC UA data structure according to the specific model and configuration of the equipment.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A BIM-based optimized management system for hoisting construction supervision, characterized in that, include: Perception Layer: Deploy 5G+ edge computing nodes to collect multi-source heterogeneous data from millimeter-wave radar, UWB+ lidar, micro lidar weather stations, and fiber optic grating sensor networks in real time. Use Kalman filter-particle filter fusion algorithm to construct a dynamic obstacle spatiotemporal map and align the spatiotemporal data of multiple sensors. Simultaneously, embed fiber optic grating sensor networks and deploy digital twin mirror interfaces to complete real-time monitoring of strain, tilt, and steel structure stress of key crane components and synchronize the BIM model with the physical entity status. Decision layer: The bottom layer uses the incremental RRT# algorithm, combined with a non-uniform sampling strategy and intelligent pruning mechanism, to generate candidate paths that satisfy the kinematic constraints of the crane, including the following steps; Initialization: Set the starting point and target point in the BIM model; Non-uniform sampling: Based on the risk heatmap, sampling is performed in key areas, with sampling probability... Based on the risk value of location x and obstacle density Calculation, the calculation formula is expressed as follows Where Ω is the sampling space, and the integral term is used to normalize the probability; Path expansion: Using the incremental RRT# algorithm, the path is randomly expanded from the starting point, and the kinematic constraints of the crane are considered in each expansion; Intelligent pruning: Removes redundant paths and retains paths with lower cost and that satisfy kinematic constraints; Path optimization: Smoothing the preserved paths; Output candidate paths: Output candidate paths that satisfy all constraints; The upper layer embeds a distributed Q-learning framework, where each crane acts as an independent intelligent agent. Through collaborative learning via a shared experience pool, it dynamically optimizes path planning strategies based on environmental perception data and construction progress requirements, including the following steps: Initialization: The agent obtains the current state information from the BIM model and initializes the Q-value table; Experience collection: Each agent collects experience during the interaction process and stores it in a shared experience pool; Collaborative learning: The agent randomly selects a batch of experiences from the experience pool for learning, and updates the Q-value table using the Q-learning algorithm based on the selected experiences; Path planning: Based on the learned Q-value table, the agent selects the optimal action for path planning; Dynamic adjustment: During the hoisting of large components of the unit, the intelligent agent selects the optimal action based on environmental perception data and construction progress requirements. The intelligent agent executes the selected action, updates its own position, and receives reward signals from the environment based on the results of the action. Simultaneously, a fuzzy sliding mode control algorithm was developed, which, combined with an LSTM-Transformer model of crane trolley and crane travel speeds, hook lifting speeds, and wire rope disturbance prediction, calculates crane motion compensation parameters. Model predictive control is then introduced, and optimal control commands are generated in advance based on predicted travel, lifting speeds, and wire rope disturbance data. This includes the following steps: Fuzzy sliding mode control algorithm: by defining sliding mode surface functions The sliding surface function represents the deviation between the system state and the desired state. ,in, It is a positional or angular error. It is the rate of change of error. These are the design parameters, and the design of the control law. The control law is expressed as follows: The system state reaches the sliding surface within a finite time and moves along the sliding surface to the equilibrium point. ,in, It is an equivalent control item. It switches control items and uses fuzzy logic to adjust the gain of the switching control items to adapt to different operating conditions; Construct an LSTM-Transformer model to predict the travel speed of the crane trolley and crane body, the lifting speed of the hook, and the wire rope disturbance: The LSTM layer processes time-series data to capture short-term changes in travel speed, lifting or lowering speed, and wire rope disturbance; the Transformer encoder processes spatial data to capture the spatial correlation between the travel speed, lifting speed, and wire rope disturbance at different locations. The model is trained using historical crane travel speed, lifting speed, and wire rope disturbance data, and the loss function is optimized to enable the model to predict the lifting data of a certain large component in the future. Integrated Model Predictive Control: Based on the LSTM-Transformer crane trolley and crane travel speed, hook lifting speed and wire rope disturbance prediction model, rolling optimization and feedback correction, the current lifting speed data is input and the predicted values ​​of future lifting speed and wire rope disturbance are output. In each control cycle, the finite-time domain optimal control problem is solved. The optimal control problem is solved using optimization algorithms to obtain the optimal control sequence. The first element of the optimal control sequence is applied to the crane system to generate the optimal control command in advance. Execution layer: The path planning instructions and wind disturbance compensation control instructions generated by the decision layer are sent to the construction equipment via the OPC UA over TSN protocol; Optimization layer: Construct a BIM model, integrate geometric information, construction progress, structural response and equipment status, and feed back the equipment execution results and structural safety monitoring data to the decision-making layer.

2. The BIM-based hoisting construction supervision and optimization management system according to claim 1, characterized in that, The multi-source heterogeneous data collected by the perception layer includes environmental perception data, structural perception data, equipment status data, and personnel and safety data. The environmental perception data includes millimeter-wave radar data, UWB+ lidar data, and micro lidar weather station data. The structural perception data includes fiber optic grating sensor network data and digital twin mirror interface data. The equipment status data includes crane operation status data and elevator operation status data. The personnel and safety data includes personnel positioning data and safety monitoring video data.

3. The BIM-based hoisting construction supervision and optimization management system according to claim 1, characterized in that, The perception layer employs a Kalman filter-particle filter fusion algorithm to perform the following specific steps for constructing a dynamic obstacle spatiotemporal map and aligning multi-sensor data spatiotemporally: Data preprocessing: Through data cleaning and data format standardization, outliers are detected and removed using thresholding or statistical methods, and a unified data structure is defined to perform preliminary processing on the raw sensor data, remove noise and outliers, and convert data from different sensors into a unified format; Kalman filtering: It recursively estimates the state of a linear Gaussian system by using the state transition matrix and the observation matrix through two steps: prediction and update. Particle filtering: It represents the state distribution of a system through a set of random samples and approximates the true posterior distribution through resampling and weight updates, thus completing the state estimation of a nonlinear non-Gaussian system. Fusion: The state estimation of the linear part is performed using Kalman filtering, and the state estimation of the nonlinear part is performed using particle filtering. The results of the two filters are then fused using a weighted average or an interactive multi-model algorithm.

4. The BIM-based hoisting construction supervision and optimization management system according to claim 1, characterized in that, The specific steps for the perception layer to perform digital twin mirror synchronization are as follows: Sensor deployment: Deploy fiber Bragg grating sensor networks on critical components of the crane; Integration of the digital twin mirror interface with the BIM model: Assign a unique identifier to each component of the BIM model, and also assign a unique identifier to each sensor in the sensor network. By establishing a mapping relationship between the identifiers, the sensor data is associated with the BIM model components. When the sensor data changes, the data is synchronized to the BIM model through the interface, driving the update of the information of the model components.

5. The BIM-based hoisting construction supervision and optimization management system according to claim 1, characterized in that, The process of sending data to construction equipment via the OPC UA over TSN protocol includes the following steps: Instruction Encoding and Encapsulation: Encode the path planning results and compensation parameters into an OPC UA data structure; OPC UA over TSN network transmission: Construct a construction equipment information model based on OPC UA, define the nodes and methods of equipment, sensors, and actuators, configure the time synchronization and traffic scheduling parameters of the TSN network, and send the encoded instructions to the construction equipment through the OPC UA over TSN protocol; Equipment-side command decoding and execution: An OPC UA client is set up on the construction equipment to receive and decode commands from the decision-making level. The OPC UA client parses the command parameters in the OPC UA data structure according to the specific model and configuration of the equipment.

Images