Electromechanical equipment operation control method and system

Abstract

The invention provides an electromechanical equipment operation control method and system, and relates to the field of electromechanical equipment operation control, and the method comprises the steps: scanning the surface of a workpiece through laser radar arrays disposed at two sides of a conveying belt, and obtaining three-dimensional point cloud data; and on the basis of space matching of the data and a preset model, real-time pose parameters of the workpiece in a global coordinate system are independently calculated, the real-time pose parameters are input into a control system after time synchronization with motion state information of an execution component of the welding robot, and the robot is driven to dynamically correct the welding track according to the actual pose. Position deviation caused by conveyor belt mechanical characteristics, power grid fluctuation and the like is avoided, the problem that the theoretical position and the real position are not consistent is solved, correction is timely and accurate, the defects such as pseudo soldering and weld penetration are completely eradicated, the product quality and the structural strength are improved, and reliable guarantee is provided for high-precision collaborative operation.

Description

Technical Field

[0001] This application relates to the field of electromechanical equipment operation control, and more specifically, to an electromechanical equipment operation control method and system. Background Technology

[0002] In the field of intelligent manufacturing, especially on production lines involving precise collaborative operations, ensuring the smooth and accurate operation of electromechanical equipment is crucial. Traditional control methods often reveal problems such as response delays, energy consumption fluctuations, and insufficient coordination between multiple devices when faced with millisecond-level response requirements, complex dynamic load changes, and precise coordination among multiple devices. These issues can seriously affect the continuity of production and the quality of products.

[0003] On automotive assembly lines, long-distance conveyor belt systems carry workpieces and work collaboratively with welding robots. Traditional control systems typically rely on sensor data (such as encoder data) from the conveyor belt drive end to calculate the theoretical position of the workpiece and use this to drive the welding robot to execute a preset welding trajectory. However, in real factory environments, instantaneous fluctuations in the mains voltage can cause abnormal torque in the conveyor belt drive motor, thus affecting the actual operating speed of the conveyor belt. Due to the inherent elasticity, transmission backlash, and significant inertial effects of the conveyor belt's mechanical structure, unpredictable instantaneous deviations can occur between the motion state of the drive wheels reflected by the sensor data from the drive end and the actual physical position of the workpiece-carrying station.

[0004] This instantaneous deviation causes a discrepancy between the theoretical workpiece position calculated by the central controller based on sensor data from the drive end and the workpiece's actual physical position. The welding robot performs welding tasks based on this erroneous theoretical position information, causing the weld point to deviate from the intended position, resulting in quality defects such as incomplete welds and burn-through, severely impacting product quality and structural strength. More importantly, this "ghost" error, generated by the combined effects of external interference and the complex dynamic response within the mechanical system, occurs randomly and is difficult to predict and reproduce. Traditional control systems, lacking independent, real-time verification methods for the workpiece's true physical position, cannot detect and correct such errors, thus failing to guarantee the accuracy of high-precision collaborative operations. Summary of the Invention

[0005] This application provides a method and system for controlling the operation of electromechanical equipment, aiming to solve the problems of response delay, energy consumption fluctuation and insufficient coordination between equipment in traditional control methods when facing complex dynamic load changes and precise coordination between multiple devices on intelligent manufacturing production lines. In particular, it addresses the problem that when a conveyor belt system and a welding robot work together, the actual physical position of the workpiece does not match the theoretical position due to external interference and complex dynamic response within the mechanical system, which leads to welding quality defects that are difficult to predict and correct.

[0006] On one hand, this application provides a method for controlling the operation of electromechanical equipment, used to perform operations on workpieces on a production line including a conveyor belt system and welding robot actuators, comprising: The surface of the workpiece is scanned by lidar arrays deployed on both sides of the conveyor belt system to obtain three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system. Based on the spatial matching relationship between the three-dimensional point cloud data and the preset workpiece model, the real-time pose parameters of the workpiece in the global coordinate system are calculated, wherein the pose parameter calculation process is independent of the sensor data at the conveyor belt drive end. The current motion state information of the welding robot's execution component is obtained, and the real-time pose parameters of the workpiece are synchronized with the current motion state information of the welding robot's execution component in time. The real-time pose parameters of the workpiece are input into the welding robot control system to drive the welding robot to dynamically correct the welding trajectory according to the actual pose of the workpiece.

[0007] On the other hand, this application provides a system for operating on workpieces on a production line including a conveyor system and welding robot actuators, the system comprising: The workpiece information acquisition module is used to scan the surface of the workpiece body by laser radar arrays deployed on both sides of the conveyor belt system, and acquire three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system. The real-time pose parameter generation module is used to calculate the real-time pose parameters of the workpiece in the global coordinate system based on the spatial matching relationship between the three-dimensional point cloud data and the preset workpiece model. The pose parameter calculation process is independent of the sensor data at the conveyor belt drive end. The motion state synchronization module is used to acquire the current motion state information of the welding robot's execution component and to synchronize the real-time pose parameters of the workpiece with the current motion state information of the welding robot's execution component in time. The compensation and correction module is used to input the real-time pose parameters of the workpiece into the welding robot control system, so as to drive the welding robot to dynamically correct the welding trajectory according to the actual pose of the workpiece.

[0008] This application relates to a method and system for controlling the operation of electromechanical equipment. By deploying lidar arrays on both sides of a conveyor belt system, it can scan the surface of the workpiece in real time and independently, acquiring high-precision three-dimensional point cloud data. Based on the spatial matching relationship between this point cloud data and a preset workpiece model, this method can accurately calculate the real-time pose parameters of the workpiece in the global coordinate system. This calculation process is independent of the sensor data from the conveyor belt drive end. This innovative design effectively avoids the instantaneous deviation between the sensor data from the drive end and the actual physical position of the workpiece caused by factors such as the inherent elasticity of the conveyor belt's mechanical structure, transmission clearance, huge inertial effects, and power grid voltage fluctuations in traditional control systems. By synchronizing the real-time pose parameters with the current motion state information of the welding robot's execution components and inputting it into the welding robot control system, the welding robot can be driven to dynamically correct the welding trajectory according to the actual pose of the workpiece. This method can effectively cope with randomly occurring "ghost" errors, improving the robustness and adaptability of the production line and providing reliable technical support for high-precision collaborative operations, thereby overcoming the shortcomings of existing technologies in detecting and correcting such errors. Attached Figure Description

[0009] To illustrate this application more clearly, the accompanying drawings used in the embodiments will be briefly described below. Obviously, those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0010] Figure 1 The diagram above illustrates a flowchart of a method for controlling the operation of electromechanical equipment. Figure 2 The diagram above illustrates a structural schematic of an electromechanical equipment operation control system.

[0011] Reference numerals: 100, Electromechanical equipment operation control system; 10, Workpiece information acquisition module; 20, Real-time pose parameter generation module; 30, Motion state synchronization module; 40, Compensation and correction module. Detailed Implementation

[0012] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0013] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0014] In the field of intelligent manufacturing, especially on production lines involving precise collaborative operations, ensuring the smooth and accurate operation of electromechanical equipment is crucial. Traditional control methods often reveal problems such as response delays, energy consumption fluctuations, and insufficient coordination between multiple devices when faced with millisecond-level response requirements, complex dynamic load changes, and precise coordination among multiple devices. These issues can severely impact production continuity and product quality. For example, suppose a long-distance conveyor belt system carries workpieces on an automotive assembly line and works collaboratively with a welding robot. Traditional control systems typically rely on sensor data (such as encoder data) from the conveyor belt drive end to calculate the theoretical position of the workpiece and use this to drive the welding robot to execute a preset welding trajectory. However, in real factory environments, instantaneous fluctuations in the mains voltage can cause abnormal torque in the conveyor belt drive motor, thus affecting the actual operating speed of the conveyor belt. Due to the inherent elasticity, transmission backlash, and significant inertial effects of the conveyor belt's mechanical structure, the motion state of the drive wheels reflected by the sensor data at the drive end will produce unpredictable instantaneous deviations from the actual physical position of the workpiece-carrying station. If the aforementioned problems are not addressed, this instantaneous deviation will cause a discrepancy between the theoretical position of the workpiece calculated by the central controller based on sensor data from the drive end and the workpiece's actual physical position. The welding robot, performing welding tasks based on erroneous theoretical position information, will cause the weld point to deviate from the predetermined position, resulting in quality defects such as incomplete welds and burn-through, severely impacting product quality and structural strength. More importantly, this "ghost" error, generated by the combined effects of external interference and the complex dynamic response within the mechanical system, occurs randomly and is difficult to predict and reproduce. Traditional control systems, lacking independent, real-time means to verify the workpiece's true physical position, cannot detect and correct such errors, thus failing to guarantee the accuracy of high-precision collaborative operations.

[0015] In this regard, such as Figure 1 As shown, this application proposes a method for controlling the operation of electromechanical equipment, used to perform operations on workpieces on a production line including a conveyor belt system and welding robot actuators, comprising: S10, by scanning the surface of the workpiece body with lidar arrays deployed on both sides of the conveyor belt system, three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system is obtained. S20, based on the spatial matching relationship between the three-dimensional point cloud data and the preset workpiece model, calculate the real-time pose parameters of the workpiece in the global coordinate system, wherein the pose parameter calculation process is independent of the sensor data at the conveyor belt drive end. S30, acquire the current motion state information of the welding robot's execution component, and synchronize the real-time pose parameters of the workpiece with the current motion state information of the welding robot's execution component in time; S40, the real-time position and posture parameters of the workpiece are input to the welding robot control system to drive the welding robot to dynamically correct the welding trajectory according to the actual position and posture of the workpiece.

[0016] This application introduces an independent lidar array for real-time workpiece pose perception and combines it with a time synchronization mechanism, enabling the welding robot to dynamically correct the welding trajectory based on the actual pose of the workpiece. This effectively overcomes the welding accuracy problem caused by the dynamic error of the conveyor system in traditional methods, and significantly improves the automation level of the production line and product quality.

[0017] The electromechanical equipment operation control method proposed in this application aims to solve the problem of inaccurate workpiece positioning leading to decreased welding quality in traditional production lines. Specifically, 3D point cloud data refers to a collection of numerous discrete points obtained by scanning the surface of an object using lidar; each point contains 3D coordinate information and can be used to reconstruct the object's surface morphology. The preset workpiece model refers to a precise 3D digital model of the workpiece pre-established before production, serving as a benchmark for comparison and positioning. Real-time pose parameters are the workpiece's position (X, Y, Z coordinates) and orientation (rotation angles around the X, Y, Z axes) relative to the global coordinate system at a given moment. The global coordinate system is the unified reference coordinate system for the entire production line or workshop, under which the poses of all equipment and workpieces are described. The welding robot actuator refers to the end effector in the welding robot that actually performs the welding operation, such as a welding torch. Current motion state information includes real-time motion data such as the speed, acceleration, and joint angles of the welding robot actuator.

[0018] The core of the electromechanical equipment operation control method of this application lies in ensuring the precise operation of the welding robot on the moving workpiece through multi-sensor fusion and real-time pose correction.

[0019] First, regarding the feature of "scanning the surface of the workpiece body by means of a LiDAR array deployed on both sides of the conveyor belt system to obtain three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system," this step is fundamental to obtaining the real-time position and attitude of the workpiece. One implementation method is to symmetrically deploy multiple LiDAR sensors on both sides of the conveyor belt system, for example, two LiDARs on each side, forming a four-sensor array. These LiDARs can perform continuous scanning using a fixed scanning frequency and field of view. As the workpiece passes through the scanning area, each LiDAR independently acquires point cloud data within its field of view. Subsequently, these independent point cloud data can be integrated into a unified local coordinate system through a pre-calibrated sensor coordinate system transformation relationship to form preliminary three-dimensional point cloud data of the workpiece. Another implementation method is to use single-line or multi-line LiDARs to scan the workpiece surface point-by-point or line-by-line through mechanical rotation or electronic scanning, thereby constructing the workpiece's three-dimensional point cloud data. For example, a lidar can be set to scan at a fixed angle and speed. When a workpiece passes by, the lidar's scanning line will cover the workpiece surface, generating a series of two-dimensional scanning data. This two-dimensional data is then integrated with the time series and the workpiece's moving speed to form a three-dimensional point cloud.

[0020] Secondly, regarding the feature of "calculating the real-time pose parameters of the workpiece in the global coordinate system based on the spatial matching relationship between the 3D point cloud data and the preset workpiece model, wherein the calculation process of the pose parameters is independent of the sensor data at the conveyor belt drive end," after acquiring the 3D point cloud data of the workpiece, it is necessary to compare it with the preset workpiece model to determine the precise pose of the workpiece. One implementation method is to directly input the acquired 3D point cloud data into a matching module based on the Iterative Closest Point (ICP) algorithm. This module will attempt to iteratively align the point cloud data with the preset workpiece model, calculating the optimal spatial transformation matrix by minimizing the distance error between the point cloud and the model. This transformation matrix contains the real-time position and orientation information of the workpiece in the global coordinate system. For example, a 3D database containing the workpiece CAD model can be pre-established. When the LiDAR acquires the workpiece point cloud data, the system will retrieve the corresponding workpiece model from the database and use the ICP algorithm for matching. During the matching process, the algorithm will continuously adjust the translation and rotation of the point cloud data relative to the model until the overlap between the two reaches a preset threshold. The final translation and rotation parameters are the real-time pose parameters of the workpiece. It is important to emphasize that the calculation of these pose parameters relies entirely on the workpiece surface data acquired by the LiDAR, and does not depend on the encoder or other sensor data at the conveyor belt drive end, thus avoiding the influence of the conveyor belt system's own errors on the pose calculation.

[0021] Secondly, regarding the feature of "acquiring the current motion state information of the welding robot's execution components and synchronizing the real-time pose parameters of the workpiece with the current motion state information of the welding robot's execution components," ensuring time consistency of information between the robot and the workpiece is crucial for collaborative operation. One implementation method is to read the motion state information of the welding robot's execution components, such as joint angles, end effector speed, and acceleration, in real time through the robot controller's internal interface. This information is typically updated at a fixed frequency (e.g., every millisecond). Simultaneously, the LiDAR system adds a timestamp after calculating the workpiece's real-time pose parameters. For time synchronization, a high-precision clock synchronization protocol (such as NTP or PTP) can be used to synchronize the time of the LiDAR system and the robot controller. For example, when the LiDAR system acquires and calculates the workpiece's pose parameters P1 at time T1, the robot controller acquires its motion state information S1 at time T1. By comparing and calibrating the timestamps, it is ensured that the pose parameters P1 and the motion state information S1 correspond in time, thus avoiding deviations in control commands due to data delays.

[0022] Finally, regarding the feature of "inputting the real-time pose parameters of the workpiece to the welding robot control system to drive the welding robot to dynamically correct the welding trajectory according to the actual pose of the workpiece," this is a crucial step in achieving precise welding. One implementation method is to send the time-synchronized real-time pose parameters of the workpiece to the welding robot control system via a high-speed communication interface (such as EtherCAT or Profinet). Upon receiving these parameters, the robot control system compares them with a preset welding trajectory. For example, if the preset welding trajectory is planned based on an ideal workpiece pose, but the actual workpiece pose deviates, the robot control system will calculate a compensation amount for the welding trajectory in real time based on this deviation. This compensation amount can be a fine-tuning of position or a correction of posture. The robot controller will dynamically adjust the motion commands of the welding robot's execution components based on the calculated compensation amount, enabling its welding torch to accurately track the actual welding path of the workpiece.

[0023] The electromechanical equipment operation control method of this application, through lidar arrays deployed on both sides of the conveyor belt system, can acquire three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system in real time and independently. This data is used to accurately calculate the real-time pose parameters of the workpiece in the global coordinate system, and this calculation process is completely independent of the sensor data at the conveyor belt drive end, thus effectively avoiding the instantaneous deviation between the theoretical position and the actual physical position of the workpiece caused by the dynamic errors of the conveyor belt system itself (such as elastic deformation, transmission backlash, inertial effects, and abnormal motor torque) in traditional methods. Subsequently, the system acquires the current motion state information of the welding robot's execution components and performs high-precision time synchronization between the real-time pose parameters of the workpiece and the robot's motion state information, ensuring the consistency of the two information on the time axis. Finally, the time-synchronized real-time pose parameters of the workpiece are input to the welding robot control system. Based on these precise real-time pose parameters, the robot control system dynamically corrects its preset welding trajectory, enabling the welding robot's execution components to accurately track the actual welding path of the workpiece, thereby avoiding quality defects such as incomplete welds and burn-through. The entire process forms a closed-loop control system, from workpiece pose perception to robot trajectory correction, realizing high-precision and adaptive welding operations on moving workpieces, significantly improving the automation level of the production line and product quality.

[0024] The electromechanical equipment operation control method of this application represents a significant technological advancement and innovation compared to traditional control systems. Traditional systems primarily rely on sensor data from the conveyor belt drive end to calculate the workpiece position. This method, when faced with inherent mechanical errors in the conveyor belt system and external interference, cannot accurately reflect the true physical position of the workpiece, leading to deviations between the welding robot's trajectory and actual requirements, thus affecting product quality. The core innovation of this application lies in the introduction of an independent laser radar array, enabling direct, real-time, and high-precision measurement of the workpiece's pose, entirely independent of the sensor data from the conveyor belt drive end. This independence completely severs the transmission path of internal errors in the conveyor belt system to the robot control system. By synchronizing the real-time pose parameters acquired by the laser radar with the motion state information of the welding robot's execution components, this application ensures that the robot control system obtains the most accurate and timely workpiece position information. Based on this, the robot control system can dynamically correct the welding trajectory, enabling the welding torch to precisely follow the actual movement of the workpiece, thereby effectively solving the welding quality problem caused by "ghost" errors in traditional methods. This control strategy, based on independent external sensing and real-time feedback correction, significantly improves the accuracy and robustness of collaborative operation of electromechanical equipment, providing reliable technical support for high-precision production in the field of intelligent manufacturing.

[0025] In some embodiments described above in this application, a scheme is proposed to input the real-time pose parameters of the workpiece into the welding robot control system to drive the welding robot to dynamically correct the welding trajectory according to the actual pose of the workpiece. However, in its implementation, if correction is based solely on simple pose deviations, it may not fully consider the motion accuracy limitations of the welding robot itself and the tolerance of actual operation for pose accuracy, which may lead to unnecessary frequent corrections, reduced production efficiency, or in some cases, failure to achieve the expected welding quality even after correction.

[0026] In response, this application further proposes a step for inputting the real-time pose parameters of the workpiece into the welding robot control system to drive the welding robot to dynamically correct the welding trajectory according to the actual pose of the workpiece, including: The real-time pose parameters of the workpiece are input into the welding robot control system; The real-time pose parameters of the workpiece are compared with the preset target position planned by the welding robot's execution component for the ideal workpiece pose. It is determined whether the comparison result meets the preset correction conditions, and a compensation command for adjusting the motion of the welding robot's execution component is generated based on the comparison result. The correction conditions are determined based on the dynamic coupling relationship between the workpiece pose offset and the welding robot's motion accuracy. The welding robot's execution components are driven to correct its welding trajectory according to the compensation command.

[0027] Specifically, the real-time pose parameters of the workpiece in the global coordinate system, calculated by the real-time pose parameter generation module, are transmitted to the welding robot control system via a data interface or communication protocol. These real-time pose parameters may include the workpiece's position information (e.g., X, Y, Z coordinates) and attitude information (e.g., rotation angles around the X, Y, Z axes).

[0028] The actual pose of the workpiece acquired at the current moment is compared with the corresponding target point pose on the pre-planned welding trajectory for the workpiece under ideal conditions, stored within the welding robot control system. This preset target position represents the ideal welding point position under conditions without pose deviation.

[0029] In practical applications, the comparison result is used to determine whether it meets the preset correction conditions, and compensation instructions are generated based on the comparison results to adjust the motion of the welding robot's actuators. The purpose is to intelligently determine when and how to perform trajectory correction. The correction conditions are determined based on the dynamic coupling relationship between the workpiece pose offset and the welding robot's motion accuracy. Specifically, the workpiece pose offset refers to the deviation between the workpiece's real-time pose parameters and the preset target position, while the welding robot's motion accuracy refers to the positioning and repeatability accuracy that the welding robot's actuators can achieve during motion. This dynamic coupling relationship means that the correction condition is not a fixed threshold, but is dynamically adjusted according to the magnitude of the workpiece pose offset and the welding robot's own motion accuracy characteristics. For example, when the workpiece pose offset is small and within the allowable range of the welding robot's motion accuracy, correction may not be triggered; however, when the offset exceeds this allowable range, correction is triggered. Furthermore, this coupling relationship can also consider the specific accuracy requirements of the welding task; for example, different welding paths or welding different materials may have different requirements for pose accuracy. Based on this judgment result, if the correction conditions are met, a compensation instruction is generated. This instruction contains specific parameters for adjusting the current motion trajectory of the welding robot's execution parts, such as displacement and rotation angle.

[0030] Therefore, after receiving the compensation command, the welding robot control system will adjust the motion planning of the welding robot's execution components in real time, so that it can make precise trajectory correction according to the actual position and posture of the workpiece when performing welding operations, ensuring that the welding tools (such as welding guns) are always in the optimal welding position and posture.

[0031] The above technical solution can significantly improve the intelligence and efficiency of electromechanical equipment operation control. This solution effectively avoids unnecessary trajectory corrections, reduces ineffective movements of the welding robot, thereby extending equipment lifespan and reducing operating costs. Furthermore, because the correction conditions are more closely aligned with actual production needs and equipment performance, the accuracy and quality of welding operations are further guaranteed, improving the overall stability and reliability of the production line.

[0032] For example, suppose a workpiece moves along a conveyor belt system on a production line and is welded by a welding robot. First, a LiDAR array scans the workpiece surface to acquire its 3D point cloud data and calculates the workpiece's real-time pose parameters in the global coordinate system. Simultaneously, the current motion state information of the welding robot's actuator is acquired and synchronized in time. Then, the workpiece's real-time pose parameters are input to the welding robot control system. This system compares the workpiece's real-time pose parameters with the preset target position planned by the welding robot for the ideal workpiece pose. For example, if the ideal welding point is at (X0, Y0, Z0), and the corresponding point detected in real-time is at (X1, Y1, Z1), the pose offset is calculated. The system then determines whether this offset meets a preset correction condition. This correction condition is not a fixed value but is determined based on the dynamic coupling relationship between the workpiece pose offset and the welding robot's motion accuracy. For example, if the welding robot's repeatability is ±0.1 mm and the allowable pose deviation for the current welding task is ±0.2 mm, then when the workpiece pose offset is less than 0.1 mm, the system may determine that no correction is needed; when the offset is between 0.1 mm and 0.2 mm, the system may generate a small compensation command based on the dynamic coupling relationship; and when the offset is greater than 0.2 mm, a more significant compensation command will be generated. Ultimately, the welding robot's execution components dynamically adjust their welding trajectory according to the generated compensation commands to ensure welding quality.

[0033] In some embodiments, the step of scanning the surface of the workpiece body by laser radar arrays deployed on both sides of the conveyor belt system to obtain three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system can be further refined, including: The workpiece is scanned synchronously by at least three sets of laser radars arranged in an interlaced pattern, each set containing two symmetrically arranged scanning units. The original point cloud obtained by scanning is filtered for reflection intensity to remove noise points with reflection intensity below a set threshold; Multi-radar point cloud data are fused into three-dimensional point cloud data in a unified coordinate system.

[0034] Specifically, at least three sets of lidar are deployed in a staggered manner on both sides of the conveyor belt system. Each set of lidar includes two symmetrically arranged scanning units configured to simultaneously scan the workpiece moving on the conveyor belt system. This layout aims to ensure comprehensive coverage of the workpiece surface, reduce blind spots, and improve the efficiency and completeness of data acquisition.

[0035] The process of filtering the reflection intensity of the raw point cloud obtained from the scan to remove noise points with reflection intensity below a set threshold can be understood as preprocessing the raw 3D point cloud data obtained after LiDAR scanning. This process analyzes the reflection intensity information of each data point, identifying and removing data points with reflection intensity below the preset threshold as noise. Its purpose is to remove low-quality or invalid data caused by environmental interference, poor workpiece surface characteristics, or sensor noise, thereby improving the purity and reliability of the point cloud data.

[0036] In practical applications, local point cloud data acquired from different LiDAR scanning units are integrated into a unified global coordinate system through coordinate transformation and registration algorithms. For example, the Iterative Closest Point (ICP) algorithm or its variants can be used, combined with the calibration parameters of each LiDAR, to accurately align and stitch point cloud data from different perspectives. The aim is to integrate scattered local data into a complete and continuous 3D point cloud model of the workpiece, providing a consistent and comprehensive data foundation for subsequent pose parameter calculations.

[0037] The above technical solutions significantly improve the quality and completeness of acquiring 3D point cloud data of moving workpieces. The staggered synchronous scanning design of multiple LiDAR arrays ensures high coverage and high-density data acquisition of the workpiece surface, reducing scanning blind spots caused by workpiece movement or complex geometry. Reflection intensity filtering effectively suppresses the impact of environmental noise and low-quality reflection points on data accuracy, improving the reliability of the point cloud data. Finally, through multi-radar data fusion, high-precision and high-completeness 3D point cloud data of the workpiece in a unified coordinate system is obtained, providing a solid data foundation for subsequent real-time pose parameter calculations, thereby ensuring the accuracy and robustness of the welding robot's dynamic correction of the welding trajectory.

[0038] In some embodiments described above, this application proposes scanning the surface of a workpiece using lidar arrays deployed on both sides of a conveyor belt system to acquire 3D point cloud data of the workpiece surface moving on the conveyor belt system. The raw point cloud data is then filtered for reflection intensity, and the multi-liquidity point cloud data is fused into 3D point cloud data in a unified coordinate system. However, in actual production environments, the workpiece moves continuously on the conveyor belt system, and the workpiece size, the conveyor belt speed, and the lidar's scanning characteristics (such as frame rate and field of view) may all change. Without effective scanning planning and control, some workpiece areas may not be scanned, or the scan data may be discontinuous or incomplete, affecting the accuracy and reliability of subsequent pose parameter calculations. Therefore, this application further proposes a scheme for dynamically planning and controlling the lidar array scanning process before acquiring 3D point cloud data to ensure continuous and comprehensive scanning of the moving workpiece.

[0039] Before the step of scanning the surface of the workpiece body by laser radar arrays deployed on both sides of the conveyor belt system to obtain three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system, the method further includes: Based on the maximum outline size of the preset workpiece model, the preset transmission speed of the conveyor belt system, and the scanning frame rate of the lidar, the scanning triggering sequence and field of view range of each scanning unit in the lidar array are dynamically planned. Based on the scanning triggering timing and field of view range of dynamic programming, the lidar array is controlled to continuously scan the moving workpiece to ensure that the workpiece is effectively covered by the field of view of at least one scanning unit at any time.

[0040] Specifically, based on the actual characteristics and operating conditions of the workpieces on the production line, the startup time and scanning coverage area of ​​each LiDAR scanning unit are intelligently adjusted. The maximum outline size of the preset workpiece model refers to the maximum external dimension of the workpiece in all dimensions, which determines the minimum scanning area that needs to be covered. The preset transmission speed of the conveyor system refers to the nominal speed at which the workpiece moves on the production line, which affects the frequency at which the scanning units need to be triggered to capture continuous data. The LiDAR's scanning frame rate refers to the number of scans the LiDAR can complete per second, which limits its data acquisition rate. By comprehensively considering these parameters, it is possible to calculate when each scanning unit should start scanning and what range its field of view should cover during workpiece movement to ensure a complete scan of the workpiece.

[0041] Furthermore, based on the scanning trigger timing and field of view range determined by dynamic programming, the lidar array is controlled to continuously scan the moving workpiece, ensuring that the workpiece is effectively covered by the field of view of at least one scanning unit at any given time. This means that the operating mode of the lidar array is adjusted in real time according to the above planning results. For example, when the workpiece enters the predetermined scanning area of ​​a scanning unit, that unit is precisely triggered to begin scanning; simultaneously, its field of view is adjusted to the optimal range to cover a specific part of the current workpiece. Through this coordinated control, even when the workpiece is moving at high speed, its surface can be effectively covered by the field of view of at least one lidar scanning unit at any given time, thereby avoiding scanning blind spots or data loss.

[0042] This application's solution, by introducing a dynamic planning step before actual scanning, can accurately calculate the optimal scanning trigger timing and field of view range in advance based on the physical dimensions of the workpiece, the conveyor belt's speed, and the performance parameters of the LiDAR itself. It is precisely this proactive planning that enables the LiDAR array to be intelligently controlled to adapt to the dynamic movement of the workpiece. By ensuring that the workpiece is effectively covered by the field of view of at least one scanning unit at any given time, this application effectively solves the scanning blind spots and data discontinuities that may exist in traditional scanning methods, thus providing a high-quality, highly complete raw data foundation for subsequent 3D point cloud data processing and pose parameter calculation.

[0043] For example, suppose a production line needs to weld different models of car bodies, which move at different speeds on a conveyor belt system. First, the system dynamically calculates and plans the optimal triggering timing and field of view for each LiDAR scanning unit based on the maximum outline dimensions of the car body model to be processed (e.g., 5 meters long, 2 meters wide, and 1.5 meters high), the real-time transmission speed of the conveyor belt system (e.g., 0.5 meters per second), and the scanning frame rate of each scanning unit in the LiDAR array (e.g., 20 frames per second). For example, for a long car body, the system might plan for multiple scanning units to be activated sequentially at different times, adjusting their field of view to form overlapping areas, ensuring continuous scanning from front to back of the car body. When the car body enters the scanning area, the control system precisely triggers the corresponding LiDAR unit to begin scanning based on the planning results, adjusting its field of view in real time to always cover the key areas of the car body. For example, when the side of the car body passes by, the side LiDAR unit is activated and adjusts its field of view to capture side data; when the top of the car body passes by, the top LiDAR unit is activated. Through this dynamic planning and control, even if the vehicle body accelerates or decelerates on the conveyor belt, or if there are significant differences in the size of different vehicle body models, the system can ensure that at any given time, every surface area of ​​the vehicle body is effectively covered by the field of view of at least one lidar scanning unit, thereby acquiring complete and high-quality three-dimensional point cloud data, providing reliable input for the subsequent precise operation of the welding robot.

[0044] In some embodiments described above, the real-time pose parameters of the workpiece in the global coordinate system are calculated based on the spatial matching relationship between 3D point cloud data and a preset workpiece model. However, in actual industrial production environments, the 3D point cloud data acquired by lidar is easily affected by various factors such as environmental noise, stray reflections, and uneven workpiece surface characteristics, leading to a decrease in point cloud data quality and consequently affecting the accuracy and reliability of pose parameter calculation. If the point cloud data quality problem is not addressed, subsequent welding robot trajectory correction may be inaccurate, affecting work quality and efficiency. Therefore, this application further proposes a method to optimize the real-time pose parameter calculation process of the workpiece in the global coordinate system. This method improves the calculation accuracy and robustness of pose parameters by performing quality assessment and anti-interference processing on the 3D point cloud data.

[0045] In some embodiments, the step of calculating the real-time pose parameters of the workpiece in the global coordinate system based on the spatial matching relationship between the three-dimensional point cloud data and the preset workpiece model includes: The three-dimensional point cloud data is subjected to data quality assessment to generate confidence information reflecting the reliability of the three-dimensional point cloud data; wherein, the data quality assessment is based on the spatiotemporal distribution characteristics of point cloud reflection intensity.

[0046] Based on the confidence information, the three-dimensional point cloud data is subjected to anti-interference processing to suppress environmental interference signals; The processed 3D point cloud data is matched with a preset workpiece model, and the real-time pose parameters of the workpiece are calculated by using a spatial transformation matrix.

[0047] Specifically, after acquiring the 3D point cloud data of the workpiece surface moving on the conveyor belt system, the first step is to perform a data quality assessment on this raw point cloud data. Data quality assessment refers to the quantitative analysis of the accuracy, completeness, consistency, and degree of influence from noise and interference on the point cloud data. Through assessment, confidence information reflecting the reliability of the 3D point cloud data can be generated. This confidence information can be a numerical value, a level, or a probability distribution, used to indicate the reliability of each point or point cloud region. The data quality assessment is based on the spatiotemporal distribution characteristics of point cloud reflection intensity. Point cloud reflection intensity refers to the intensity of the reflected signal received by the lidar, which is related to factors such as the surface material, color, distance of the measured object, and laser incident angle. Spatiotemporal distribution characteristics refer to the spatial distribution pattern and temporal trend of reflection intensity. For example, abnormally high or low reflection intensity, drastic changes in reflection intensity within a short period, or irregular spatial distribution of reflection intensity may indicate quality problems or interference in the point cloud data. By analyzing these characteristics, abnormal points or interfered areas can be effectively identified.

[0048] Furthermore, after generating confidence information, the 3D point cloud data undergoes anti-interference processing based on this confidence information. The purpose of anti-interference processing is to suppress environmental interference signals, such as reflections from other light sources, sensor noise, or accidental reflections from other equipment on the production line. Confidence information plays a crucial role in this process; for example, it can be used to weight, filter, remove, or correct point cloud data with low confidence. For instance, points with low confidence can have their weight reduced in subsequent matching algorithms, or they can be directly treated as noise points and removed. In this way, the impact of interference signals on pose calculation can be effectively reduced, improving the purity of the point cloud data.

[0049] Finally, the anti-interference processed 3D point cloud data is matched with a preset workpiece model, and the real-time pose parameters of the workpiece are calculated using a spatial transformation matrix. The preset workpiece model is a CAD model or high-precision scanned model of the workpiece, containing precise geometric information. The matching process aims to find the optimal spatial correspondence between the processed point cloud data and the preset workpiece model. The spatial transformation matrix is ​​a 4x4 homogeneous matrix containing rotation and translation information, used to describe the transformation relationship of the point cloud data from its current coordinate system to the coordinate system of the preset workpiece model (or the global coordinate system). By solving this matrix, the real-time position and orientation of the workpiece in the global coordinate system, i.e., the real-time pose parameters, can be accurately determined.

[0050] Through the above technical solution, this application can significantly improve the calculation accuracy and robustness of the real-time pose parameters of the workpiece. Compared with directly using raw point cloud data for matching, this application effectively filters out environmental noise and interference signals through data quality assessment and anti-interference processing, avoiding pose calculation errors caused by point cloud data quality issues. This enables the welding robot to dynamically correct the welding trajectory based on a more accurate actual workpiece pose, thereby improving the accuracy and quality of welding operations, reducing the scrap rate, and enhancing the overall automation level and reliability of the production line.

[0051] For example, suppose that on a production line, when the surface of a workpiece moving on a conveyor belt is scanned by a lidar array, occasional flashes from other equipment in the production workshop or abnormal local reflective properties of the workpiece surface can cause some of the acquired 3D point cloud data to contain noise points with high reflectivity or abnormal distribution. According to the solution in this application, the original 3D point cloud data is first assessed for data quality. For example, by analyzing whether the reflectivity of these abnormal points is significantly higher than that of surrounding normal points, or whether they exhibit an instantaneous pulse pattern in a time series, these interfering points can be identified and assigned lower confidence information. Subsequently, based on this confidence information, the system performs anti-interference processing on the point cloud data, for example, by removing point cloud data with confidence levels below a preset threshold, or by performing weighted averaging to reduce their impact. After processing, the resulting 3D point cloud data will be cleaner and more accurately reflect the true geometry of the workpiece. Finally, these processed, high-quality 3D point cloud data are matched with a pre-defined workpiece model. Using spatial matching methods such as the Iterative Closest Point (ICP) algorithm, the precise real-time pose parameters of the workpiece in the global coordinate system are calculated. For example, the workpiece's offset relative to the conveyor belt in the X, Y, and Z directions, as well as its rotation angles around each axis, can be obtained. These precise pose parameters are then input into the welding robot control system, enabling the welding robot to adjust its welding path in real time according to the actual position and orientation of the workpiece, ensuring the accuracy and quality of the welding operation.

[0052] In some embodiments, the step of performing data quality assessment on 3D point cloud data and generating confidence information reflecting the reliability of the 3D point cloud data may include the following process: The three-dimensional point cloud data is preliminarily processed to separate different types of reflection signals; The preliminary processing aims to distinguish the valid workpiece reflection signals in the original 3D point cloud data from various potential interference signals (e.g., reflections from environmental debris, other equipment, multipath effects, or sensor noise). Specifically, preliminary screening and clustering can be performed based on multiple dimensions such as point cloud density, reflection intensity, geometric features, or temporal variations, thereby dividing reflection points with different physical or statistical characteristics into different signal clusters.

[0053] For the different types of reflected signals after separation, their respective characteristic parameters are calculated. Specifically, for each type of separated reflection signal, a series of characteristic parameters that characterize its properties are extracted. For example, for workpiece reflection signals, the average reflection intensity, point density, surface normal distribution, principal component analysis (PCA) features, etc., of its point cloud can be calculated; for potential interference signals, the range of reflection intensity fluctuations, spatial dispersion, duration, contrast with the background, etc., may be calculated. These characteristic parameters aim to quantify the unique properties of different signals.

[0054] The feature parameters are compared with the feature patterns of different preset interference sources to identify the types of interference sources currently existing. In practical applications, the system pre-establishes a database containing feature patterns of various known interference sources (such as dust, water mist, ambient light, other sensor signals, and surface reflections from conveyor belt systems). By comparing the calculated real-time feature parameters with the preset patterns in this database, pattern recognition algorithms (such as support vector machines, neural networks, decision trees, or rule-based expert systems) can be used to determine whether a specific interference source exists in the current point cloud data and identify its specific type.

[0055] Based on the identified interference source type, a targeted data quality assessment is performed on the 3D point cloud data, and confidence information reflecting the reliability of the 3D point cloud data is generated.

[0056] Therefore, once a specific type of interference source is identified, the system can conduct a targeted assessment based on the degree of impact of that interference source on the quality of the point cloud data. For example, a lower confidence level can be assigned to the corresponding point cloud data for a specific type of interference, while a higher confidence level can be assigned to clear workpiece reflections. The confidence level information can be a numerical value (e.g., a probability value between 0 and 1) used to quantify the data reliability of each point or region, providing a basis for subsequent anti-interference processing and pose calculation.

[0057] Through the above technical solution, this application can significantly improve the accuracy and precision of 3D point cloud data quality assessment. Compared with generalized assessment based solely on the spatiotemporal distribution characteristics of point cloud reflection intensity, this solution identifies and distinguishes different types of reflection signals and interference sources, making data quality assessment more targeted. Consequently, the generated confidence information can more accurately quantify the reliability of point cloud data, effectively avoiding misjudgments caused by unidentified interference sources. This provides a more reliable input for subsequent anti-interference processing, ultimately ensuring higher accuracy and stability in the calculation of real-time workpiece pose parameters, especially in complex and ever-changing industrial production environments.

[0058] In some embodiments, the step of performing preliminary processing on 3D point cloud data to separate different types of reflection signals may include the following steps: Acquire three-dimensional point cloud data of the workpiece moving on the conveyor belt system; Local features are extracted from each reflection point in the three-dimensional point cloud data; Based on a preset reflection signal feature library, the local features are initially classified, and the reflection signals are divided into multiple potential categories. The preliminary classification of reflection signals is processed by region growing or clustering to aggregate adjacent reflection points with similar characteristics into different reflection signal clusters; Based on the internal characteristic differences of the reflected signal clusters, an adaptive boundary partitioning strategy is adopted to distinguish different types of reflected signal clusters; Multi-view consistency cross-validation is performed on the reflected signal clusters to further subdivide them into subclusters with different interference source characteristics; The segmented and verified clusters of reflected signals are separated from the three-dimensional point cloud data of the main target workpiece.

[0059] Acquiring the 3D point cloud data of the workpiece moving on the conveyor belt system is the starting point of the entire processing flow, providing the raw data foundation for subsequent feature extraction and signal separation. Furthermore, for each point cloud data point, geometric, intensity, or texture attributes such as normal vector, curvature, reflection intensity value, and point density are calculated within its surrounding neighborhood. These local features effectively characterize the local geometric structure and physical properties of the reflection point, providing a basis for subsequent classification and clustering.

[0060] Based on a pre-defined feature library of reflection signals, local features are initially classified into multiple potential categories. This feature library pre-stores typical local feature patterns for different types of reflection signals (such as workpiece body, environmental clutter, sensor noise, and reflections from other objects). By comparing the extracted local features with the feature library, reflection signals from different sources that may exist in the point cloud data can be preliminarily identified.

[0061] The preliminary classification of reflected signals is processed by region growing or clustering, aggregating adjacent reflected points with similar characteristics into different reflected signal clusters. The purpose is to group reflected points with the same or similar physical origins and spatial continuity into discrete signal clusters that can be analyzed independently. Region growing algorithms can start from a seed point and gradually incorporate neighboring similar points; clustering algorithms can group points based on distance or feature similarity between them.

[0062] Specifically, based on the internal characteristic differences of reflection signal clusters, an adaptive boundary segmentation strategy is adopted to distinguish different types of reflection signal clusters. The aim is to finely identify and separate reflection signals with different properties. Internal characteristic differences can include point cloud density, reflection intensity distribution, and geometry. The adaptive boundary segmentation strategy can dynamically adjust the segmentation criteria according to these differences, avoiding misjudgments caused by fixed thresholds.

[0063] Furthermore, multi-view consistency cross-validation is performed on the reflected signal clusters to further subdivide them into subclusters with different interference source characteristics, aiming to improve the accuracy and robustness of the separation. Multi-view consistency refers to the consistency of observation results for the same reflected signal cluster from data acquired from different lidar scanning units or at different time points. Through cross-validation, the nature of the reflected signal clusters can be further confirmed, and subclusters with different interference source characteristics that are difficult to distinguish from a single viewpoint can be identified.

[0064] Finally, the segmented and verified clusters of reflection signals are separated from the 3D point cloud data of the main target workpiece. The purpose is to accurately remove all reflection signals that are not related to the workpiece body from the original point cloud data, thereby obtaining pure 3D point cloud data containing only the workpiece body information, providing high-quality input for subsequent pose calculations.

[0065] The above technical solution enables precise separation of different types of reflection signals in 3D point cloud data, significantly improving the precision of data preprocessing. Through a multi-stage, adaptive processing flow, this solution effectively suppresses the influence of environmental interference signals on the workpiece's point cloud data, thus providing a more reliable input for subsequent data quality assessment. This ensures that the generated confidence information more accurately reflects the true reliability of the 3D point cloud data, thereby improving the anti-interference processing effect based on confidence information. Ultimately, this guarantees the accuracy and stability of the workpiece's real-time pose parameter calculation, providing high-quality data support for the dynamic correction of welding trajectories by the welding robot.

[0066] In some embodiments described above, an adaptive boundary partitioning strategy is employed to distinguish different types of reflection signal clusters based on their internal characteristic differences. However, in practical applications, the internal characteristic differences of reflection signal clusters can be complex and variable. For example, when there is environmental interference or changes in workpiece surface characteristics, a simple adaptive strategy may struggle to accurately identify and partition different types of reflection signal clusters, thus affecting the accuracy of subsequent data quality assessment. If this problem is not addressed, interference signals may fail to be effectively separated, thereby affecting the accuracy of calculating the real-time pose parameters of the workpiece. Therefore, this application further proposes a more refined and robust boundary partitioning strategy, which improves the accuracy of distinguishing different types of reflection signal clusters through multi-dimensional feature analysis, dynamic threshold adjustment, predictive threshold management, and auxiliary judgment mechanisms.

[0067] In some embodiments, the step of employing an adaptive boundary partitioning strategy based on the internal characteristic differences of the reflected signal clusters to distinguish different types of reflected signal clusters includes: The local reflection intensity distribution, point cloud density gradient, and geometric features of the reflected signal cluster are analyzed to calculate the internal feature difference index of the reflected signal cluster. Based on the aforementioned internal feature difference index, a dynamic threshold adjustment mechanism is adopted to automatically adjust the boundary division threshold according to the real-time data distribution characteristics. When the internal characteristic difference index of the reflected signal cluster changes gradually, a threshold range is predicted based on historical data and current trends, and the boundary division threshold is dynamically adjusted within this range. When multiple reflected signal clusters are found to have highly similar internal feature differences, an auxiliary judgment mechanism based on spatial proximity and temporal continuity is introduced to preferentially classify the reflected signal clusters that have spatial isolation or temporal evolution patterns.

[0068] This process involves multi-dimensional feature extraction for each cluster of reflected signals. Local reflection intensity distribution can be understood as the reflected light intensity values ​​at each point in the point cloud and their spatial distribution within the cluster; point cloud density gradient refers to the spatial density of the point cloud and its changing trend; geometric features include, but are not limited to, the cluster's size, shape, convexity, concavity, linearity, and flatness. These features are comprehensively calculated to generate an index that fully reflects the internal differences within the reflected signal cluster, aiming to provide a quantitative basis for subsequent precise segmentation.

[0069] Furthermore, based on the aforementioned internal feature difference indicators, a dynamic threshold adjustment mechanism is adopted to automatically adjust the boundary division threshold according to the real-time data distribution characteristics. This means that the system does not use a fixed threshold for division, but rather adjusts the threshold used to distinguish different reflection signal clusters in real time and adaptively according to the actual feature differences of the current point cloud data. For example, when the difference indicator is large, the threshold may be relaxed; when the difference indicator is small, the threshold may be tightened, with the aim of ensuring optimal division results under scenarios of varying complexity.

[0070] Furthermore, when the system detects a gradual change in the internal characteristic difference index of the reflected signal cluster, it predicts a suitable threshold range based on historical data and current trends, and dynamically adjusts the boundary division threshold within this range. Specifically, when the characteristics of the reflected signal cluster exhibit relative stability, the system uses past experience data and current evolution trends to estimate a reasonable threshold range. Within this estimated range, the boundary division threshold is fine-tuned based on the current gradual change. The purpose is to maintain division stability while avoiding unnecessary frequent adjustments due to small fluctuations, thereby improving the robustness of the system.

[0071] Furthermore, when multiple reflection signal clusters are detected to have highly similar internal feature differences, this application introduces an auxiliary judgment mechanism based on spatial proximity and temporal continuity to prioritize the classification of reflection signal clusters with spatial isolation or temporal evolution patterns. This means that when faced with ambiguous situations where features are highly similar and difficult to distinguish directly, the system introduces additional contextual information for auxiliary decision-making. Spatial proximity refers to considering the relative positional relationship of reflection signal clusters in three-dimensional space, such as whether they are independent or closely adjacent; temporal continuity refers to observing the appearance, movement, or disappearance patterns of reflection signal clusters in continuous scan frames. Through these auxiliary judgments, the system can more intelligently identify and prioritize reflection signal clusters that are clearly isolated in space or exhibit stable evolution patterns in time. The aim is to solve complex scenarios that are difficult to distinguish based on internal features alone, and to improve the accuracy and reliability of classification.

[0072] This application's solution effectively addresses the limitations of traditional adaptive boundary partitioning strategies when dealing with complex and variable reflection signal clusters by introducing multi-dimensional feature analysis, dynamic threshold adjustment, predictive threshold management, and auxiliary judgment mechanisms. Specifically, by analyzing the local reflection intensity distribution, point cloud density gradient, and geometric features of reflection signal clusters, the internal feature differences can be quantified more comprehensively and accurately, providing a solid data foundation for subsequent partitioning. Based on this, a dynamic threshold adjustment mechanism is employed, allowing the boundary partitioning threshold to be adaptively adjusted according to real-time data distribution characteristics, avoiding the limitations of fixed thresholds in different scenarios. Furthermore, when the internal feature difference indicators of reflection signal clusters change gradually, the threshold range is predicted by combining historical data and current trends and dynamically adjusted, ensuring the stability and accuracy of partitioning and avoiding misjudgments due to minor fluctuations. In addition, for complex situations where the internal feature difference indicators of multiple reflection signal clusters are highly similar, an auxiliary judgment mechanism based on spatial proximity and temporal continuity is introduced. This mechanism can utilize contextual information for more intelligent decision-making, prioritizing the partitioning of reflection signal clusters with clear spatial or temporal patterns, thereby significantly improving the discrimination accuracy and robustness of different types of reflection signal clusters.

[0073] For example, suppose on a welding production line, the surface of a workpiece moving on a conveyor belt system has slight oil stains or reflective spots, and there is a small amount of spattered welding slag nearby. These factors may produce clusters of reflected signals that are similar to but distinct from the reflected signals of the workpiece itself during lidar scanning.

[0074] First, the system analyzes the local reflection intensity distribution, point cloud density gradient, and geometric features of these reflected signal clusters. For example, oil stains may result in a relatively uniform but slightly lower reflection intensity distribution, while welding slag may exhibit high density and small-sized geometric features. Based on these analyses, an internal feature difference index is calculated.

[0075] Next, the system dynamically adjusts the boundary segmentation threshold based on these differences. If the characteristics of oil stains and welding slag differ significantly from those of the workpiece body, the threshold will be widened for faster separation; if the differences are small, the threshold will be tightened for more precise differentiation.

[0076] Furthermore, if the internal feature difference index of a certain cluster of reflection signals (e.g., slight reflection from the workpiece surface) changes slowly, the system will combine historical data (e.g., the performance of this type of reflection on different workpieces) and the current trend to predict a suitable threshold range, and fine-tune the boundary division threshold within this range to ensure stable and accurate separation from the main workpiece point cloud data.

[0077] Finally, when multiple clusters of reflected signals (e.g., weld slag and oil contaminants from different locations) exhibit highly similar internal characteristic differences, the system introduces an auxiliary judgment mechanism. For example, weld slag typically has a fixed spatial location (e.g., at the edge of the workpiece or in a specific area) and its appearance exhibits temporal continuity (e.g., persisting for a period of time after welding). Oil contaminants, on the other hand, may be randomly distributed on the workpiece surface. By analyzing these spatial proximity and temporal continuity, the system can prioritize the identification and segmentation of weld slag clusters with clear spatial isolation or temporal evolution patterns. Even if their internal characteristics are temporarily similar to those of oil contaminant clusters, confusion can be avoided, thus ensuring accurate identification and separation of different interference sources.

[0078] In some embodiments described above in this application, when the internal characteristic difference index of the reflected signal cluster changes gradually, although the threshold range can be predicted and dynamically adjusted based on historical data and current trends, in practical applications, simple prediction and adjustment may not be sufficient to cope with the complex and ever-changing characteristics of the reflected signal cluster. Especially in scenarios requiring high-precision segmentation, there may be issues with insufficiently fine threshold adjustment or slow response, thus affecting the accuracy of data quality assessment. Therefore, this application further proposes a scheme for more refined adjustment of the boundary segmentation threshold to improve the accuracy of data quality assessment and the robustness of anti-interference processing.

[0079] When the internal characteristic difference index of the reflected signal cluster changes gradually, the step of predicting a threshold range based on historical data and current trends, and dynamically adjusting the boundary division threshold within this range, includes: The local reflection intensity distribution, point cloud density gradient, and geometric features of the reflected signal cluster are analyzed in real time to obtain the current rate of change and direction of the reflected signal cluster. By combining the threshold interval distribution of reflected signal clusters with similar rates and directions of change in historical data, the initial prediction threshold interval is determined; Within the initial prediction threshold range, the boundary division threshold is fine-tuned based on the proximity of the internal characteristic difference index of the current reflected signal cluster to the threshold range boundary, as well as the rate and direction of change. The threshold range is shrunk or expanded based on the fine-tuning results.

[0080] Specifically, by continuously monitoring changes in these characteristic parameters, such as calculating their first or second derivatives, the evolution trend of the reflection signal cluster over time can be quantified. This allows us to obtain the current rate and direction of change of the reflection signal cluster; for example, whether the reflection intensity is gradually increasing or decreasing, whether the point cloud density is becoming more concentrated or dispersed, and whether the geometry remains stable or undergoes deformation.

[0081] Specifically, by combining the threshold interval distribution of reflected signal clusters with similar rates and directions of change in historical data, the initial prediction threshold interval is determined. This can be understood as the system maintaining a historical database that records the effective boundary division of threshold intervals corresponding to the internal characteristic difference indicators of reflected signal clusters under different rates and directions of change. When a new reflected signal cluster appears, its current rate of change and direction are compared with historical data to match the closest historical pattern, and a preliminary, empirical threshold interval is determined as the initial prediction threshold interval.

[0082] In practical applications, the real-time state of the current reflected signal cluster is further considered based on the initial prediction interval. For example, if the internal feature difference index is very close to a certain boundary, and the rate of change indicates that it is moving towards that boundary, the threshold can be more aggressively fine-tuned in that direction to ensure the accuracy of the segmentation. This fine-tuning mechanism makes threshold adjustment more flexible and precise, and can adapt to subtle changes.

[0083] Furthermore, the threshold interval can be narrowed or expanded based on the fine-tuning results. The purpose is to optimize subsequent threshold prediction and adjustment processes. If the fine-tuned threshold exhibits good segmentation and is located in the central region of the current threshold interval, the interval can be appropriately narrowed to improve the accuracy of future predictions. Conversely, if the fine-tuned threshold is close to the edge of the interval or the segmentation effect is poor, it may be necessary to expand the threshold interval to increase adjustment flexibility and avoid missing the optimal threshold due to an overly narrow interval.

[0084] Through the above technical solution, this application can adjust the boundary segmentation threshold more precisely and intelligently, especially in complex scenarios where the internal feature difference index of reflection signal clusters changes gradually. This significantly improves the accuracy of 3D point cloud data quality assessment, ensuring that different types of reflection signal clusters can be effectively identified and separated even in subtly changing environments. Therefore, environmental interference signals can be suppressed more effectively, providing higher-quality input data for subsequent workpiece pose parameter calculations, thereby improving the accuracy and reliability of the entire electromechanical equipment operation control method.

[0085] For example, suppose that on the surface of a workpiece moving on a conveyor belt system, due to slight changes in material or lighting, the local reflection intensity distribution, point cloud density gradient, and internal variability indices of geometric features of a certain cluster of reflected signals exhibit a slow and continuous decreasing trend.

[0086] First, the system analyzes the local reflection intensity distribution, point cloud density gradient, and geometric features of the reflected signal cluster in real time, and calculates its current rate of change and direction. For example, the intensity decrease rate is 0.01 units per second, and the density gradient change rate is 0.005 units per second / pixel.

[0087] Next, the system queries the historical database to find historical records with similar descent rates and directions to the current reflected signal cluster. Assuming historical data shows that under similar trends, the effective boundary delimitation threshold range is typically between 0.45 and 0.55, the system determines 0.45-0.55 as the initial prediction threshold range.

[0088] Within this initial prediction threshold range, the system further fine-tunes the boundary division threshold based on the proximity of the internal characteristic difference index of the current reflected signal cluster (e.g., the current value is 0.48) to the threshold range boundary, and its continuous downward trend. For example, since the current value of 0.48 is closer to the lower boundary of 0.45 and the trend is downward, the system may fine-tune the threshold to 0.47 to identify potential boundary changes earlier.

[0089] Finally, based on the results of this fine-tuning, if the threshold of 0.47 performs well in actual segmentation, the system may narrow the threshold range to 0.46-0.50 to optimize future prediction accuracy. Through this dynamic and fine-tuning process, the accuracy and adaptability of boundary segmentation can be ensured even when feature changes are gradual.

[0090] like Figure 2 As shown in the illustration, a schematic diagram of the structure of an electromechanical equipment operation control system is also provided as an example of a specific embodiment of this application. Specifically, an electromechanical equipment operation control system 100 is used to perform operations on workpieces on a production line including a conveyor belt system and welding robot execution components. The system includes: The workpiece information acquisition module 10 is used to scan the surface of the workpiece body by laser radar arrays deployed on both sides of the conveyor belt system, and acquire three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system. The real-time pose parameter generation module 20 is used to calculate the real-time pose parameters of the workpiece in the global coordinate system based on the spatial matching relationship between the three-dimensional point cloud data and the preset workpiece model. The pose parameter calculation process is independent of the sensor data at the conveyor belt drive end. The motion state synchronization module 30 is used to acquire the current motion state information of the welding robot's execution component and to synchronize the real-time pose parameters of the workpiece with the current motion state information of the welding robot's execution component in time. The compensation and correction module 40 is used to input the real-time position and orientation parameters of the workpiece into the welding robot control system, so as to drive the welding robot to dynamically correct the welding trajectory according to the actual position and orientation of the workpiece.

[0091] This system, through its modular design, achieves precise, real-time control of moving workpieces on the production line. The workpiece information acquisition module, independent of the conveyor belt system, directly senses the surface information of the workpiece using a lidar array, thus avoiding the inaccurate workpiece positioning problems caused by the dynamic errors of the conveyor belt system itself, a problem common in traditional methods. The real-time pose parameter generation module calculates the workpiece's true pose in the global coordinate system based on the acquired high-precision 3D point cloud data. Subsequently, the motion state synchronization module ensures the time consistency between the workpiece pose information and the motion state information of the welding robot's actuators, laying the foundation for subsequent precise control. Finally, the compensation and correction module inputs these precise and synchronized pose parameters into the welding robot control system, enabling the welding robot to dynamically adjust its welding trajectory according to the actual pose of the workpiece. This significantly improves welding accuracy and product quality, effectively solving quality defects such as incomplete welds and burn-through caused by workpiece positioning errors in traditional production lines.

[0092] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method of controlling operation of an electromechanical device for performing work on a workpiece on a production line including a conveyor system and a welding robot execution unit, characterized by, Comprising: acquiring three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system by scanning the workpiece body surface through the laser radar array deployed on both sides of the conveyor belt system; calculating the real-time pose parameters of the workpiece in the global coordinate system based on the spatial matching relationship between the three-dimensional point cloud data and the preset workpiece model, wherein the pose parameter calculation process is independent of the sensor data of the conveyor belt driving end; acquiring the current motion state information of the welding robot execution component, and time synchronizing the real-time pose parameters of the workpiece with the current motion state information of the welding robot execution component; inputting the real-time pose parameters of the workpiece into the welding robot control system to drive the welding robot to dynamically correct the welding trajectory according to the actual pose of the workpiece.

2. The electromechanical equipment operation control method according to claim 1, characterized by, The step of inputting the real-time pose parameters of the workpiece into the welding robot control system to drive the welding robot to dynamically correct the welding trajectory according to the actual pose of the workpiece includes: inputting the real-time pose parameters of the workpiece into the welding robot control system; comparing the real-time pose parameters of the workpiece with the preset target position planned by the welding robot execution component for the ideal workpiece pose, judging whether the comparison result satisfies the preset correction condition, and generating compensation instructions for adjusting the motion of the welding robot execution component based on the comparison result; wherein the correction condition is determined based on the dynamic coupling relationship between the workpiece pose offset and the welding robot motion accuracy; driving the welding robot execution component to correct its welding trajectory according to the compensation instructions.

3. The electromechanical equipment operation control method according to claim 1, characterized by, The step of acquiring three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system by scanning the workpiece body surface through the laser radar array deployed on both sides of the conveyor belt system includes: synchronously scanning the workpiece through at least three groups of laser radars arranged in a staggered manner, each group containing two symmetrically arranged scanning units; performing reflection intensity filtering on the original point cloud obtained by scanning to remove noise points with reflection intensity lower than a set threshold; fusing the multi-radar point cloud data into three-dimensional point cloud data in a unified coordinate system.

4. The electromechanical equipment operation control method according to claim 3, characterized by, Before the step of acquiring three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system by scanning the workpiece body surface through the laser radar array deployed on both sides of the conveyor belt system, it further includes: dynamically planning the scanning trigger timing and field of view angle range of each scanning unit in the laser radar array according to the maximum contour size of the preset workpiece model, the preset transmission speed of the conveyor belt system, and the scanning frame rate of the laser radar; based on the dynamically planned scanning trigger timing and field of view angle range, controlling the laser radar array to continuously scan the moving workpiece to ensure that the workpiece is effectively covered by the field of view of at least one scanning unit at any time.

5. The electromechanical equipment operation control method according to claim 1, characterized by, The step of calculating the real-time pose parameters of the workpiece in the global coordinate system based on the spatial matching relationship between the three-dimensional point cloud data and the preset workpiece model includes: performing data quality evaluation on the three-dimensional point cloud data to generate confidence information reflecting the reliability of the three-dimensional point cloud data; wherein the data quality evaluation is realized based on the spatiotemporal distribution characteristics of point cloud reflection intensity; Based on the confidence information, the three-dimensional point cloud data is subjected to anti-interference processing to suppress environmental interference signals; Based on the processed three-dimensional point cloud data, a preset workpiece model is matched, and a real-time pose parameter of the workpiece is solved through a spatial transformation matrix.

6. The electromechanical equipment operation control method according to claim 5, characterized by, The step of performing data quality assessment on the three-dimensional point cloud data to generate confidence information reflecting the reliability of the three-dimensional point cloud data comprises: The three-dimensional point cloud data is subjected to preliminary processing to separate different types of reflection signals; The feature parameters of the separated different types of reflection signals are calculated respectively; The feature parameters are compared with preset feature modes of different interference sources to identify the type of the current interference source; According to the identified type of the interference source, the three-dimensional point cloud data is subjected to targeted data quality assessment, and confidence information reflecting the reliability of the three-dimensional point cloud data is generated.

7. The electromechanical equipment operation control method according to claim 6, characterized by, The step of performing preliminary processing on the three-dimensional point cloud data to separate different types of reflection signals comprises: Obtain the three-dimensional point cloud data of the workpiece moving on the conveyor belt system; Extract local features for each reflection point in the three-dimensional point cloud data; According to a preset reflection signal feature library, the local features are preliminarily classified, and the reflection signals are divided into multiple potential categories; The preliminarily classified reflection signals are subjected to region growing or clustering processing, and adjacent reflection points with similar features are aggregated into different reflection signal clusters; According to the internal feature difference of the reflection signal cluster, an adaptive boundary division strategy is adopted to distinguish different types of the reflection signal cluster; Multi-view consistency cross-validation is performed on the reflection signal cluster to subdivide sub-clusters with different interference source features; The divided and verified reflection signal cluster is separated from the three-dimensional point cloud data of the main target workpiece.

8. The electromechanical equipment operation control method according to claim 7, characterized by, The step of adopting an adaptive boundary division strategy according to the internal feature difference of the reflection signal cluster to distinguish different types of the reflection signal cluster comprises: Analyze the local reflection intensity distribution, point cloud density gradient and geometric shape feature of the reflection signal cluster to calculate the internal feature difference index of the reflection signal cluster; Based on the internal feature difference index, a dynamic threshold adjustment mechanism is adopted to automatically adjust the boundary division threshold according to real-time data distribution characteristics; When it is detected that the internal feature difference index of the reflection signal cluster changes gently, the threshold interval is predicted according to historical data and current trend, and the boundary division threshold is dynamically adjusted within the interval; When it is detected that the internal feature difference indexes of multiple reflection signal clusters are highly similar, an auxiliary judgment mechanism based on spatial proximity and time continuity is introduced to preferentially divide the reflection signal clusters with spatial isolation or time evolution law.

9. The electromechanical equipment operation control method according to claim 8, characterized by, The step of predicting the threshold interval according to historical data and current trend when it is detected that the internal feature difference index of the reflection signal cluster changes gently, and dynamically adjusting the boundary division threshold within the interval comprises: Real-time analysis is performed on the local reflection intensity distribution, point cloud density gradient, and geometric shape features of the reflection signal cluster to obtain a current change rate and direction of the reflection signal cluster; An initial prediction threshold interval is determined in combination with a threshold interval distribution of reflection signal clusters with similar change rates and directions in historical data; Within the initial prediction threshold interval, a boundary division threshold is fine-tuned according to the proximity of an internal feature difference index of the current reflection signal cluster to the boundary of the threshold interval, as well as the change rate and direction; The threshold interval is contracted or expanded according to the fine-tuning result.

10. An electromechanical equipment operation control system for performing operations on workpieces on a production line including a conveyor belt system and welding robot actuators, characterized in that, The system comprises: A workpiece information acquisition module for scanning the surface of a workpiece body through a laser radar array deployed on both sides of a conveyor belt system to acquire three-dimensional point cloud data of the workpiece surface moving on the conveyor belt system; A real-time pose parameter generation module for calculating real-time pose parameters of the workpiece in a global coordinate system based on the spatial matching relationship between the three-dimensional point cloud data and a preset workpiece model, wherein the pose parameter calculation process is independent of sensor data of a conveyor drive end; A motion state synchronization module for acquiring current motion state information of a welding robot execution component and time-synchronizing the real-time pose parameters of the workpiece and the current motion state information of the welding robot execution component; A compensation correction module for inputting the real-time pose parameters of the workpiece into a welding robot control system to drive the welding robot to dynamically correct a welding trajectory according to the actual pose of the workpiece.

Images