Construction machinery collision safety protection method and system

Abstract

The invention is suitable for the technical field of construction safety, and provides a construction machinery collision safety protection method and system, and the method comprises the steps: constructing a distributed communication network, sharing the state information of the pose, speed and the like of a machine in real time, and dynamically managing the network topology; the binocular camera, the millimeter wave radar and the IMU are fused to perceive the environment, and the operation path of the adjacent machine in the next five seconds is predicted; a cooperative control strategy of mechanical arm kinematics constraint and sensor view field coverage is established, the track is optimized, and the pointing direction of the sensor is adjusted; and executing three-level collision risk protection, and triggering early warning, potential field correction and emergency shutdown according to the collision time TTC. According to the scheme, full-process intelligentization from information interaction to risk protection is achieved through multi-technology fusion, the safety and cooperation efficiency of the construction machinery under complex working conditions are improved, and the method is suitable for multi-machinery operation scenes in the fields of buildings, traffic and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of construction safety, and in particular relates to a construction machinery collision safety protection method and system. Background Art

[0002] With the increasing scale and complexity of infrastructure construction, the number of collaborative operations involving construction machinery in areas such as construction, transportation, and energy is increasing. Intensive, automated operations involving cranes, excavators, and loaders have become the norm. Against this backdrop, collision safety protection technology for construction machinery faces significant challenges. On the one hand, the need for multiple machines to collaboratively share state information, such as posture and motion parameters, in real time places higher demands on the stability of the communication network and the efficiency of data exchange. On the other hand, the accuracy of environmental perception and the ability to predict operational intent in complex working conditions (such as dust, rain, fog, and confined spaces) become key factors limiting safety protection effectiveness. While existing technologies have gradually introduced sensor monitoring and simple communication capabilities, technical bottlenecks remain in core areas such as multi-device collaborative perception, dynamic trajectory planning, and hierarchical collision protection.

[0003] The closest existing technology currently uses sensors (such as cameras and radars) deployed on a single machine for local environmental monitoring, combined with simple wireless communications to achieve limited sharing of status data between devices. Some solutions attempt to build environmental models using single-modality sensor data (such as relying solely on vision or radar) and trigger braking responses based on fixed safety distance thresholds. In terms of communication architecture, single-band wireless communication technologies (such as Wi-Fi or Bluetooth) are mostly used, and there is a lack of unified standards for data formats. At the risk protection level, a "warning-shutdown" binary strategy is generally adopted, and a refined response mechanism for collision risk levels has not been formed. In addition, existing solutions do not coordinate the optimization of sensor layout and robot arm motion trajectory, resulting in monitoring blind spots during complex motion processes. The prediction of adjacent machine operating intentions relies solely on historical trajectories or single control instructions, lacking the deep reasoning capabilities of multimodal data fusion. Summary of the Invention

[0004] The purpose of the present invention is to provide a construction machinery collision safety protection method, aiming to solve the technical problems existing in the prior art identified in the background technology.

[0005] The present invention is achieved by providing a construction machinery collision safety protection method, the method comprising:

[0006] Build a distributed communication network for construction machinery groups. Use a hybrid communication architecture to collect real-time status information for each construction machine and share it across the group. Define a standardized data format containing machine position and motion parameters, and implement dynamic topology management.

[0007] Using a multimodal sensor array to perceive the surrounding environment, collect visual data, radar data, and inertial measurement data, establish a multi-device collaborative perception model, and predict the operating intention of each adjacent machine;

[0008] Based on the perception results of the multi-device collaborative perception model, a collaborative control strategy for each robot's motion trajectory and sensor layout is generated. A mechanical coupling model that considers the robot's kinematic constraints is established to optimize motion path planning and adjust the spatial orientation of the camera and sensor in real time.

[0009] Implement multi-level dynamic protection against collision risks, establish a layered warning threshold mechanism, and use the safety potential field algorithm to correct the equipment's motion trajectory.

[0010] As a further solution of the present invention, the construction of a distributed communication network for a group of construction machines collects the real-time status information of each construction machine through a hybrid communication architecture and shares it among the group of construction machines, specifically including:

[0011] Build a dual-band communication network including 5G / Wi-Fi6 main channel and LoRa auxiliary channel;

[0012] Define the data packet structure, including the device unique identifier, three-dimensional coordinates, attitude quaternion, velocity vector and trajectory prediction coefficient;

[0013] The neighbor discovery protocol is used to dynamically maintain device connection status and update the network topology with a period of 100ms.

[0014] As a further solution of the present invention, the method of sensing the surrounding environment through a multimodal sensor array, collecting visual data, radar data, and inertial measurement data, establishing a multi-device collaborative perception model, and predicting the operating intention of each adjacent machine specifically includes:

[0015] Read sensor data from the binocular camera, millimeter-wave radar, and IMU installed on the machine body;

[0016] Construct a 3D point cloud map of the environment by fusing multi-source sensor data through spatiotemporal alignment algorithms;

[0017] A deep learning model is used to analyze adjacent machine control instructions and historical trajectories, and predict the operation path in the next 5 seconds.

[0018] As a further solution of the present invention, the deep learning model is used to analyze the adjacent mechanical control instructions and historical trajectories to predict the operation path in the next 5 seconds. The specific formula is:

[0019] ;

[0020] in, Indicates that the machine The location at the moment, is the control point of the third-order Bezier curve, is the dynamic correction value of the trajectory starting point, is the instantaneous motion direction vector, is the task constraint point, It is the target position prediction after 5 seconds, specifically expressed as:

[0021] ;

[0022] The three-dimensional trajectory sequence of adjacent machines in the past 5 seconds, is the current control instruction vector, is the activation function, is the weight matrix, is the bias term.

[0023] As a further solution of the present invention, the generation of a coordinated control strategy for the motion trajectory and sensor layout of each robot, the establishment of a mechanical coupling model considering the kinematic constraints of the robot arm, the optimization of motion path planning, and the real-time adjustment of the spatial orientation of the camera and sensor specifically include:

[0024] Establish a mathematical model of the robot arm joint motion and sensor field of view coverage, and define the coverage optimization objective function;

[0025] Embed sensor coverage constraints in the RRT* path planning algorithm to generate collision-free motion trajectories;

[0026] Control the electric pan / tilt to adjust the camera's pitch and azimuth angles to maintain a field of view overlap rate of ≥25% in the dangerous area.

[0027] As a further solution of the present invention, the implementation of multi-level dynamic collision risk protection, establishment of a layered warning threshold mechanism, and correction of the device motion trajectory in combination with a safety potential field algorithm specifically include:

[0028] Calculate the collision time TTC based on the relative speed and distance between devices and set the three-level response threshold;

[0029] When TTC is less than or equal to 5s, the safety potential field algorithm is activated to generate a path correction vector to avoid the collision risk area;

[0030] When TTC≤2s, the robot arm power output is cut off and an emergency stop command is broadcast to associated equipment.

[0031] As a further solution of the present invention, the deep learning model is a multimodal fusion network, comprising:

[0032] Trajectory analysis branch: uses bidirectional LSTM to process historical position sequences;

[0033] Instruction parsing branch: CNN is used to extract control instruction features;

[0034] Fusion decision layer: weighting each modality feature through the attention mechanism;

[0035] Output layer: Regression predicts the third-order Bezier curve control points of the future trajectory.

[0036] As a further solution of the present invention, the objective function is expressed as:

[0037] ;

[0038] in, represents the sensor coverage area, is the end speed of the robot arm, is the collision risk factor, is the weight factor, is the total number of time steps, is the current time step.

[0039] Another object of the present invention is to provide a construction machinery collision safety protection system, the system comprising:

[0040] A distributed communication network building module is used to build a distributed communication network for construction machinery groups. This module collects real-time status information of each construction machinery through a hybrid communication architecture and shares it among the construction machinery groups. It also defines a standardized data format containing machine posture and motion parameters and implements dynamic topology management.

[0041] The operation intention prediction module is used to perceive the surrounding environment through a multimodal sensor array, collect visual data, radar data, and inertial measurement data, establish a multi-device collaborative perception model, and predict the operation intention of each adjacent machine;

[0042] The motion trajectory generation module is used to generate the motion trajectory of each machine and the collaborative control strategy of the sensor layout based on the perception results of the multi-device collaborative perception model. It also establishes a mechanical coupling model that considers the kinematic constraints of the manipulator, optimizes the motion path planning, and adjusts the spatial orientation of the camera and sensor in real time.

[0043] The collision risk dynamic protection module is used to perform multi-level collision risk dynamic protection, establish a layered warning threshold mechanism, and correct the equipment motion trajectory in combination with the safety potential field algorithm.

[0044] The beneficial effects of the present invention are:

[0045] The present invention realizes efficient sharing of real-time status information of multiple machines by constructing a distributed communication network. The dual-band hybrid architecture and dynamic topology management ensure the stability and real-time performance of communication in complex environments. The multimodal sensor fusion is combined with the deep learning model to accurately predict the operating intentions of adjacent machines, transforming passive collision detection into active risk prediction. The collaborative control strategy of motion trajectory and sensor layout dynamically adjusts the monitoring perspective while optimizing the path planning of the robot arm, eliminating monitoring blind spots and improving environmental perception accuracy. The multi-level collision risk dynamic protection mechanism realizes refined response from early warning to emergency braking through hierarchical thresholds and safety potential field algorithms, which not only ensures safety but also reduces interference with operating efficiency. The overall solution forms a closed-loop system of "communication interconnection-collaborative perception-intelligent planning-hierarchical protection", which effectively solves the problems of inefficient information interaction, intention prediction deviation, trajectory conflict and collision protection lag in the collaborative operation of multiple machines, and significantly improves the safety and operating efficiency of construction machinery under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flowchart of a construction machinery collision safety protection method provided by an embodiment of the present invention;

[0047] Figure 2 A flowchart of building a distributed communication network for construction machinery groups provided by an embodiment of the present invention;

[0048] Figure 3 A flowchart for predicting the operating intention of each adjacent machine provided by an embodiment of the present invention;

[0049] Figure 4 A flowchart for optimizing motion path planning and adjusting the spatial orientation of the camera and sensor in real time, provided by an embodiment of the present invention;

[0050] Figure 5 A flow chart of correcting the motion trajectory of a device provided in an embodiment of the present invention;

[0051] Figure 6 This is a structural block diagram of the construction machinery collision safety protection system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0053] Figure 1 A flowchart of a construction machinery collision safety protection method provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes:

[0054] S100: Build a distributed communication network for construction machinery. This network uses a hybrid communication architecture to collect real-time status information for each construction machine and share it across the entire construction machinery group. This network also defines a standardized data format containing machine position and motion parameters, and implements dynamic topology management.

[0055] This step achieves the complementary advantages of different communication technologies by building a dual-band communication network consisting of 5G / Wi-Fi6 main channels and LoRa auxiliary channels. The 5G and Wi-Fi6 main channels, with their high speed and low latency, can meet the needs of real-time status information transmission, ensuring that key data such as the three-dimensional coordinates, attitude quaternions, and velocity vectors of each construction machine can be quickly shared among the group with an update cycle of 100ms, providing instant data support for real-time collaborative control.

[0056] The LoRa auxiliary channel takes advantage of its low power consumption and wide coverage to ensure communication stability in complex construction environments (such as multiple obstacles and long-distance work scenarios), avoid connection interruptions caused by main channel signal obstruction, and ensure comprehensive network coverage.

[0057] A data packet structure is defined that includes a unique device identifier, three-dimensional coordinates, attitude quaternion, velocity vector, and trajectory prediction coefficient, forming a unified standardized data format. This format not only fully describes the position and motion state of the machine, but also provides a motion trend reference for adjacent machines through the trajectory prediction coefficient, eliminating data format differences between different devices, enabling each machine to quickly parse and utilize shared information, and improving the overall data interaction efficiency of the system.

[0058] The neighbor discovery protocol is used to dynamically maintain the device connection status and update the network topology in a 100ms cycle. It can perceive the location changes and entry and exit of construction machinery in real time, adapting to the characteristics of frequent equipment movement and dynamic changes in network nodes in construction scenarios, ensuring that the communication network always maintains the optimal connection status, and avoiding the connection failure problem that occurs in the static network architecture when the equipment moves.

[0059] This step uses a hybrid architecture of a dual-band communication network to build a communication system that combines high speed, low latency, wide coverage, and high reliability. It can stably transmit real-time status information in complex construction environments, providing a solid communication foundation for collaborative operations of multiple machines.

[0060] The definition of a standardized data format breaks down information barriers between devices, enabling construction machinery from different manufacturers and types to exchange information based on a unified data language, significantly improving the efficiency and accuracy of group collaboration. The dynamic topology management mechanism empowers the communication network with dynamic adaptability, ensuring real-time network reconfiguration as equipment moves and the work area changes, maintaining efficient data transmission links and avoiding information delays or loss due to network connectivity issues. This fundamentally ensures the real-time and effectiveness of subsequent functions such as work intention prediction, trajectory planning, and collision avoidance.

[0061] For example, at large bridge construction sites, multiple cranes and pavers operate simultaneously and frequently adjust their positions. The dual-band network ensures real-time sharing of information such as the position and speed of each machine. Dynamic topology management enables new machines entering the work area to quickly access the network. Standardized data formats allow equipment from different brands to work together without additional adaptation, effectively avoiding the risk of collisions caused by communication delays or incompatible data formats, and significantly improving construction safety and operating efficiency.

[0062] like Figure 2 As shown, the construction of a distributed communication network for construction machinery groups collects real-time status information of each construction machinery through a hybrid communication architecture and shares it among the construction machinery groups, specifically including:

[0063] S110 builds a dual-band communication network with 5G / Wi-Fi6 primary channels and LoRa auxiliary channels;

[0064] S120, defining a data packet structure including a device unique identifier, three-dimensional coordinates, attitude quaternion, velocity vector, and trajectory prediction coefficients;

[0065] S130: Use the neighbor discovery protocol to dynamically maintain the device connection status and update the network topology with a period of 100ms.

[0066] The S200 uses a multimodal sensor array to perceive the surrounding environment, collect visual data, radar data, and inertial measurement data, establish a multi-device collaborative perception model, and predict the operating intention of each adjacent machine;

[0067] In this step, in the process of perceiving the surrounding environment and predicting the operating intentions of adjacent machines through a multimodal sensor array, the binocular camera, millimeter-wave radar, and IMU installed on the machine body are first used to achieve multi-dimensional real-time collection of visual images, distance and speed parameters, and attitude acceleration data.

[0068] Among them, the binocular camera can provide high-resolution visual information for identifying the outline of obstacles, the boundaries of the working area and the appearance features of other machines; the millimeter-wave radar can work stably in severe weather such as rain and fog, and accurately measure the distance, relative speed and azimuth of the target object; the IMU provides real-time feedback on the attitude angle and acceleration of the machine body, providing an inertial reference for spatial positioning.

[0069] The three types of sensor data are timestamp synchronized and coordinate converted through the spatiotemporal alignment algorithm, and the observation data of different sensors are unified into the same spatiotemporal coordinate system. A high-precision point cloud map containing the three-dimensional coordinates, reflection intensity and other information of environmental objects is constructed, providing three-dimensional environmental basic data for subsequent operation intention analysis.

[0070] On this basis, a multimodal fusion deep learning model composed of bidirectional LSTM and CNN is used to jointly analyze the three-dimensional trajectory sequence of adjacent machines in the past 5 seconds and the current control instructions: the bidirectional LSTM can capture the time dependency in the historical trajectory and explore the periodic and trend characteristics of mechanical movement (such as uniform linear motion, circular turning, etc.); CNN extracts features from the operation codes in the control instructions (such as "arm lifting" and "slewing braking") to identify the type and intensity of the current operation action.

[0071] By using the attention mechanism to weightedly fuse trajectory features and command features, the model can accurately predict the operation paths of adjacent machines within the next 5 seconds. For example, it can determine in advance whether a crane is about to perform a hoisting rotation or whether an excavator is entering the high-frequency action stage of material loading and unloading.

[0072] This prediction method, which combines historical motion patterns with current operational intentions, can more accurately capture the complex logic of mechanical operations and reduce prediction deviations caused by sudden operations, compared to models that rely solely on trajectories or instructions.

[0073] This step builds an intelligent perception system capable of understanding the environment and predicting intentions through the collaborative perception of multimodal sensors and the joint reasoning of deep models. Multi-source data fusion effectively overcomes the limitations of single sensors (such as the blind spots of visual sensors at night and radar's blind spots for detecting small obstacles), enabling the system to stably acquire surrounding environmental information in all weather conditions and complex working conditions, providing reliable data support for subsequent decision-making.

[0074] The operation intention prediction function transforms passive collision detection into active risk prediction. By predicting the movement trajectory of adjacent machinery 5 seconds in advance, the machine can adjust its own movement strategy at an early stage to avoid emergency braking or path conflicts caused by delayed reaction.

[0075] For example, in a tunnel construction scenario, when the loader predicts the excavator's intention to "move to the left front to load soil" based on the tunnel contour identified by the camera, the position of the excavator in front detected by the radar, and its own tilt angle fed back by the IMU, combined with the deep learning model, it can plan a detour route in advance to avoid scratches in narrow spaces.

[0076] This forward-looking perception capability not only improves the safety of group construction machinery operations but also indirectly increases construction efficiency by reducing ineffective waiting and emergency adjustments. Furthermore, the attention mechanism of the multimodal fusion network adaptively assigns weights to different data modalities. This ensures the robustness of the perception model in dynamically changing construction environments (for example, automatically increasing the weight of radar data when visual data noise increases due to dust occlusion). This provides stable input information for subsequent trajectory planning and collision avoidance modules, fundamentally enhancing the environmental adaptability and decision-making reliability of the entire safety protection system.

[0077] like Figure 3 As shown, the multimodal sensor array is used to perceive the surrounding environment, collect visual data, radar data, and inertial measurement data, establish a multi-device collaborative perception model, and predict the operating intention of each adjacent machine, specifically including:

[0078] S210, reading sensor data from a binocular camera, a millimeter-wave radar, and an IMU provided on the machine body;

[0079] S220, which fuses multi-source sensor data through a spatiotemporal alignment algorithm to construct a 3D point cloud map of the environment;

[0080] S230 uses a deep learning model to analyze adjacent machine control instructions and historical trajectories to predict the operation path in the next 5 seconds.

[0081] In this step, the deep learning model is used to analyze the adjacent mechanical control instructions and historical trajectories to predict the operation path in the next 5 seconds. The specific formula is:

[0082] ;

[0083] in, Indicates that the machine is The location at the moment, is the control point of the third-order Bezier curve, is the dynamic correction value of the trajectory starting point, is the instantaneous motion direction vector, is the task constraint point, It is the target position prediction after 5 seconds, specifically expressed as:

[0084] ;

[0085] The three-dimensional trajectory sequence of adjacent machines in the past 5 seconds, is the current control instruction vector, is the activation function, is the weight matrix, is the bias term.

[0086] In this step, the deep learning model is a multimodal fusion network, including:

[0087] Trajectory analysis branch: uses bidirectional LSTM to process historical position sequences;

[0088] Instruction parsing branch: CNN is used to extract control instruction features;

[0089] Fusion decision layer: weighting each modality feature through the attention mechanism;

[0090] Output layer: Regression predicts the third-order Bezier curve control points of the future trajectory.

[0091] S300, based on the perception results of the multi-device collaborative perception model, generates a collaborative control strategy for each robot's motion trajectory and sensor layout, establishes a mechanical coupling model that considers the robot's kinematic constraints, optimizes motion path planning, and adjusts the spatial orientation of the camera and sensor in real time;

[0092] In this step, a mathematical mapping relationship between the robot arm joint angle, end effector position and sensor field of view range is established, and the sensor coverage area, robot arm movement speed and collision risk coefficient are integrated into a unified objective function to form an optimization model that takes into account both environmental perception and motion planning.

[0093] Specifically, by analyzing the Denavit-Hartenberg parameters of the robotic arm, a kinematic forward and inverse solution model is constructed. Combining the perspective projection matrix of the binocular camera with the spherical coordinate system transformation of the radar, the spatial coverage of the sensor at different joint angles is determined.

[0094] On this basis, the node expansion conditions of the RRT* path planning algorithm are deeply coupled with the sensor coverage constraints. This not only requires that the generated trajectory meet the kinematic limitations of the robot arm (such as the upper limit of the joint angular velocity and the end load torque constraint), but also ensures that within each time step, the dangerous area (such as the operating radius of the robot arm and the intersection area of the motion path of the adjacent machine) is within the overlapping field of view of at least two sensors. The camera pitch and azimuth angles are adjusted in real time through the electric pan-tilt platform to ensure that the field of view overlap rate of the key monitoring area is always maintained above 25%.

[0095] This collaborative control strategy breaks the limitation of traditional path planning that only focuses on the feasibility of the robot arm's own movement, and dynamically incorporates the sensor layout into the trajectory optimization process. For example, when the robot arm performs a large-angle rotation, the system will predict the motion trajectory of the end effector in advance and synchronously drive the gimbal to ensure that obstacles on the rotation path are always within the joint monitoring range of the camera and radar, avoiding environmental perception failure caused by blind spots.

[0096] This step builds a closed-loop collaborative mechanism of "perception-planning-execution". By embedding the sensor field of view coverage requirements into the motion planning algorithm, it achieves a deep integration of the robot arm's motion trajectory and environmental perception capabilities.

[0097] Specifically, the kinematically constrained mechanical coupling model accurately describes the mapping between the robot arm's joint motion and the end-user trajectory, avoiding the risk of uncontrolled motion due to joint overruns or mechanical structure interference. This is particularly true in multi-robot collaborative operations (such as the coordination between a placing arm and a crane during concrete pouring). Trajectory optimization can be used to reserve safe movement space for adjacent robots.

[0098] On the other hand, the dynamic adjustment strategy of the sensor layout effectively solves the monitoring blind spot problem of traditional fixed-view sensors. By maintaining the field of view overlap in dangerous areas, it ensures that under complex working conditions (such as multiple equipment operating in tunnels and construction outside high-rise buildings), the system can capture the micro-movement changes of surrounding machinery and the position offset of obstacles in real time.

[0099] For example, during cantilever casting construction of a bridge, when the basket-mounted robotic arm is vibrating segmented concrete, the system will dynamically adjust the scanning frequency of the lidar and the exposure parameters of the camera according to the vibration frequency and amplitude of the robotic arm, while optimizing the vibration trajectory of the robotic arm, so that the sensor can clearly capture the concrete shape in the vibration area while avoiding trajectory planning deviations caused by mechanical vibrations, thereby improving operation accuracy and reducing collision risks by more than 60%.

[0100] In addition, the speed and collision risk weighting terms introduced in the objective function enable the system to automatically shrink the sensor coverage to focus on key areas during high-speed operations, and expand the monitoring range to prevent potential collisions during low-speed precision operations. This adaptive adjustment capability significantly enhances the safety and operating efficiency of construction machinery in multi-task switching scenarios.

[0101] like Figure 4 As shown, the collaborative control strategy for generating the motion trajectory of each machine and the sensor layout, establishing a mechanical coupling model considering the kinematic constraints of the manipulator, optimizing the motion path planning, and adjusting the spatial orientation of the camera and sensor in real time, specifically includes:

[0102] S310, establishing a mathematical model of the robot arm joint motion and sensor field of view coverage, and defining a coverage optimization objective function;

[0103] S320, embeds sensor coverage constraints in the RRT* path planning algorithm to generate collision-free motion trajectories;

[0104] S330 controls the electric pan / tilt to adjust the camera's pitch and azimuth angles to maintain a field of view overlap rate of ≥25% in the danger zone.

[0105] In this step, the objective function is expressed as:

[0106] ;

[0107] in, represents the sensor coverage area, is the end speed of the robot arm, is the collision risk factor, is the weight factor, is the total number of time steps, is the current time step.

[0108] S400: Implement multi-level dynamic protection against collision risks, establish a layered warning threshold mechanism, and use a safety potential field algorithm to correct the device's motion trajectory.

[0109] This step first calculates the time-to-collision (TTC) using the relative velocity vector and three-dimensional coordinates between devices, and constructs a three-level response system including early warning, avoidance, and braking.

[0110] Specifically, three levels of thresholds are set based on the maximum braking distance of the construction machinery, the inertial load characteristics of the robotic arm, and the complexity of the working environment:

[0111] When TTC>5s, a level 1 warning is triggered, prompting the driver to pay attention to surrounding dynamics through the on-board display and buzzer;

[0112] When 5s ≥ TTC > 2s, the secondary response is activated. The safety potential field algorithm generates a vector field containing repulsive and attractive forces based on the real-time position and predicted trajectory of the surrounding machinery. It applies a repulsive potential field to neighboring machinery and obstacles, and an attractive potential field to the safe operating area, driving the movement trajectory of the machine to shift to a low-risk area. At the same time, by dynamically adjusting the acceleration of the robot arm joints, a safe distance is gradually increased while avoiding sudden stops.

[0113] When TTC is less than or equal to 2s, the third-level emergency brake is activated, immediately cutting off the power output of the robot arm's hydraulic system, triggering the electromagnetic brake to lock the joint, and broadcasting a shutdown command to related equipment within 30 meters through the LoRa channel, forcing surrounding machinery to enter a safe standby state synchronously.

[0114] This layered mechanism breaks through the traditional binary model of collision protection, which is "no warning or shutdown". Through refined risk-grading response, it ensures safety while minimizing interference with normal operations.

[0115] For example, in a port container loading and unloading scenario, when the TTC calculated value of the reach stacker and forklift shows 4s, the safety potential field algorithm will generate a lateral offset trajectory based on the movement direction of the two, allowing the reach stacker to adjust its route without interrupting the lifting operation, avoiding the risk of container shaking caused by rigid shutdown; and when the TTC drops sharply to 1.5s, the system sends a shutdown signal to the five surrounding devices while cutting off power, effectively curbing the possibility of multi-machine chain collision.

[0116] The significant advantage of this step lies in the construction of a three-dimensional protection system of "risk classification-dynamic response-coordinated braking." Through real-time TTC calculations, a quantitative assessment of collision risk is achieved, enabling the system to dynamically adapt protection strategies based on the urgency of the danger. The introduction of a safety potential field algorithm empowers the machine to autonomously avoid collisions. The resulting continuous, smooth, and corrected trajectory not only complies with the manipulator's kinematic constraints (such as joint angular velocity limits), but also intelligently selects the optimal obstacle avoidance path through the principle of potential field superposition. Compared to the traditional hard limit of a fixed safety distance, this flexible protection mechanism can reduce ineffective downtime by over 30% in multi-machinery intensive operation scenarios.

[0117] The three-level threshold setting balances safety, reliability and operational efficiency: the first-level warning provides a buffer period for manual intervention, which is suitable for routine scenarios such as low-speed adjustments; the second-level potential field correction realizes automated risk control to avoid accidents caused by delayed driver reaction; the third-level emergency braking serves as the last safety barrier, and minimizes collision damage in extreme situations through power cut-off and global broadcast mechanisms.

[0118] For example, at a subway shield construction site, when the TTC of the shield machine's propulsion system and the synchronous grouting machinery is close to 3s, the safety potential field algorithm will generate a slight offset instruction along the tunnel axis based on the shield machine's excavation speed and the rotation radius of the grouting arm, ensuring that the two maintain a dynamic safety distance of 0.5 meters in a small space; and when a sudden hydraulic system failure causes the robotic arm to lose control and the TTC quickly drops to 1s, the system completes power cut-off within 200ms and links the surrounding equipment within 50 meters to shut down, successfully avoiding tunnel structure damage accidents caused by the robotic arm's high-speed impact on the shield segments.

[0119] This progressive protection strategy not only improves the survivability of construction machinery under complex working conditions, but also builds a regional safety protection network through a coordinated shutdown mechanism, upgrading from single-point protection to group safety collaboration, providing a revolutionary solution for equipment cluster management in high-risk working environments.

[0120] like Figure 5 As shown, the multi-level collision risk dynamic protection is implemented, a layered warning threshold mechanism is established, and the device motion trajectory is corrected in combination with the safety potential field algorithm, specifically including:

[0121] S410, calculating the collision time TTC based on the relative speed and distance between the devices and setting the three-level response threshold;

[0122] S420, when TTC ≤ 5s, activate the safety potential field algorithm to generate a path correction vector to avoid the collision risk area;

[0123] S430: When TTC is less than or equal to 2s, the robot arm power output is cut off and an emergency stop command is broadcast to associated equipment.

[0124] Figure 6 The structural block diagram of the construction machinery collision safety protection system provided by the embodiment of the present invention is as follows: Figure 6 As shown, the system includes:

[0125] The distributed communication network construction module 100 is used to build a distributed communication network for a group of construction machines. It collects real-time status information of each construction machine through a hybrid communication architecture and shares it among the group of construction machines. It defines a standardized data format containing machine posture and motion parameters and implements dynamic topology management.

[0126] The operation intention prediction module 200 is used to perceive the surrounding environment through a multimodal sensor array, collect visual data, radar data, and inertial measurement data, establish a multi-device collaborative perception model, and predict the operation intention of each adjacent machine;

[0127] The motion trajectory generation module 300 is used to generate the motion trajectory of each robot and the coordinated control strategy of the sensor layout based on the perception results of the multi-device collaborative perception model, establish a mechanical coupling model that considers the kinematic constraints of the robot arm, optimize the motion path planning, and adjust the spatial orientation of the camera and sensor in real time;

[0128] The collision risk dynamic protection module 400 is used to perform multi-level collision risk dynamic protection, establish a layered warning threshold mechanism, and correct the device motion trajectory in combination with a safety potential field algorithm.

[0129] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0131] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A construction machinery collision safety protection method, characterized in that: The method comprises: Build a distributed communication network for construction machinery groups. Use a hybrid communication architecture to collect real-time status information for each construction machine and share it across the group. Define a standardized data format containing machine position and motion parameters, and implement dynamic topology management. Using a multimodal sensor array to perceive the surrounding environment, collect visual data, radar data, and inertial measurement data, establish a multi-device collaborative perception model, and predict the operating intention of each adjacent machine; Based on the perception results of the multi-device collaborative perception model, a collaborative control strategy for each robot's motion trajectory and sensor layout is generated. A mechanical coupling model that considers the robot's kinematic constraints is established to optimize motion path planning and adjust the spatial orientation of the camera and sensor in real time. Implement multi-level dynamic protection against collision risks, establish a layered warning threshold mechanism, and use the safety potential field algorithm to correct the equipment's motion trajectory.

2. The method according to claim 1, characterized in that The construction of a distributed communication network for construction machinery groups collects real-time status information of each construction machinery through a hybrid communication architecture and shares it among the construction machinery groups. Specifically, the following steps are involved: Build a dual-band communication network including 5G / Wi-Fi6 main channel and LoRa auxiliary channel; Define the data packet structure, including the device unique identifier, three-dimensional coordinates, attitude quaternion, velocity vector and trajectory prediction coefficient; The neighbor discovery protocol is used to dynamically maintain device connection status and update the network topology with a period of 100ms.

3. The method according to claim 1, characterized in that The multimodal sensor array is used to perceive the surrounding environment, collect visual data, radar data, and inertial measurement data, establish a multi-device collaborative perception model, and predict the operating intention of each adjacent machine. Specifically, it includes: Read sensor data from the binocular camera, millimeter-wave radar, and IMU installed on the machine body; Construct a 3D point cloud map of the environment by fusing multi-source sensor data through spatiotemporal alignment algorithms; A deep learning model is used to analyze adjacent machine control instructions and historical trajectories, and predict the operation path in the next 5 seconds.

4. The method according to claim 3, characterized in that The deep learning model is used to analyze the adjacent mechanical control instructions and historical trajectories to predict the operation path in the next 5 seconds. The specific formula is: ； in, Indicates that the machine is The location at the moment, is the control point of the third-order Bezier curve, is the dynamic correction value of the trajectory starting point, is the instantaneous motion direction vector, is the task constraint point, It is the target position prediction after 5 seconds, specifically expressed as: ； The three-dimensional trajectory sequence of adjacent machines in the past 5 seconds, is the current control instruction vector, is the activation function, is the weight matrix, is the bias term.

5. The method according to claim 1, wherein The above mentioned process generates a coordinated control strategy for each robot's motion trajectory and sensor layout, establishes a mechanical coupling model that considers the robot's kinematic constraints, optimizes motion path planning, and adjusts the spatial orientation of the camera and sensor in real time. Specifically, it includes: Establish a mathematical model of the robot arm joint motion and sensor field of view coverage, and define the coverage optimization objective function; Embed sensor coverage constraints in the RRT* path planning algorithm to generate collision-free motion trajectories; Control the electric pan / tilt to adjust the camera's pitch and azimuth angles to maintain a field of view overlap rate of ≥25% in the dangerous area.

6. The method according to claim 1, characterized in that The implementation of multi-level dynamic collision risk protection, establishment of a layered warning threshold mechanism, and correction of the device motion trajectory in combination with a safety potential field algorithm specifically include: Calculate the collision time TTC based on the relative speed and distance between devices and set the three-level response threshold; When TTC is less than or equal to 5s, the safety potential field algorithm is activated to generate a path correction vector to avoid the collision risk area; When TTC≤2s, the robot arm power output is cut off and an emergency stop command is broadcast to associated equipment.

7. The method according to claim 3, characterized in that The deep learning model is a multimodal fusion network, including: Trajectory analysis branch: uses bidirectional LSTM to process historical position sequences; Instruction parsing branch: CNN is used to extract control instruction features; Fusion decision layer: weighting each modality feature through the attention mechanism; Output layer: Regression predicts the third-order Bezier curve control points of the future trajectory.

8. The method according to claim 4, characterized in that The objective function is expressed as: ； in, represents the sensor coverage area, is the end speed of the robot arm, is the collision risk factor, is the weight factor, is the total number of time steps, is the current time step.

9. Construction machinery collision safety protection system, characterized by: The system comprises: A distributed communication network building module is used to build a distributed communication network for construction machinery groups. This module collects real-time status information of each construction machinery through a hybrid communication architecture and shares it among the construction machinery groups. It also defines a standardized data format containing machine posture and motion parameters and implements dynamic topology management. The operation intention prediction module is used to perceive the surrounding environment through a multimodal sensor array, collect visual data, radar data, and inertial measurement data, establish a multi-device collaborative perception model, and predict the operation intention of each adjacent machine; The motion trajectory generation module is used to generate the motion trajectory of each machine and the collaborative control strategy of the sensor layout based on the perception results of the multi-device collaborative perception model. It also establishes a mechanical coupling model that considers the kinematic constraints of the manipulator, optimizes the motion path planning, and adjusts the spatial orientation of the camera and sensor in real time. The collision risk dynamic protection module is used to perform multi-level collision risk dynamic protection, establish a layered warning threshold mechanism, and correct the equipment motion trajectory in combination with the safety potential field algorithm.

Images