A gear precision assembly system and method based on visual servoing

Abstract

This invention discloses a vision servo-based precision gear assembly system and method, belonging to the field of automated assembly technology. The system includes a vision perception layer, a decision planning layer, and an execution control layer. The vision perception layer employs a global shutter monocular camera and a ring-shaped shadowless light source, deploying a dual-fusion recognition algorithm based on edge gradients and grayscale histograms, and setting a grayscale difference threshold to suppress illumination and reflection interference. The decision planning layer executes adaptive hand-eye calibration, segmented vision servo trajectory planning, and high-frequency closed-loop control; the segmented trajectory includes a rapid movement segment, a transition segment, and a precision assembly segment, with speed and feedback frequency matched hierarchically. Hole coordinates and gear angles are simultaneously calculated through a single image capture, achieving integrated pose assembly; the precision assembly segment constructs a 10ms-level high-frequency closed loop, with PID dynamic path correction. This invention achieves a recognition success rate ≥98%, 3-hole assembly ≤3 minutes, accuracy ≤0.02mm, high pass rate, and fast changeover, making it suitable for precision assembly applications.

Description

Technical Field

[0001] This invention relates to the field of automated assembly technology, and specifically to a vision servo-based precision gear assembly system and method. Background Technology

[0002] In the field of mechanical engineering, gear assembly is a core and critical step in the manufacturing of transmission systems. Its assembly precision directly determines the operational stability, noise control, transmission efficiency, and service life of the transmission system. As industrial automation develops towards higher precision, higher flexibility, and higher cycle time, traditional manual assembly and fixed-track robot assembly can no longer meet the precision requirements of gear assembly in high-end manufacturing. Vision-guided robot servo assembly technology has become the mainstream development direction in the industry.

[0003] Monocular vision has become a common vision solution for gear assembly in industrial settings due to its advantages of low hardware cost, flexible deployment, and strong field adaptability. However, existing monocular vision-guided gear assembly technology still has several core shortcomings, and it is still unable to balance assembly accuracy, production efficiency, operational stability, and scene adaptability.

[0004] Firstly, at the visual recognition level, existing solutions generally adopt single-feature recognition schemes. Monocular vision is easily affected by sudden changes in workshop lighting and interference from reflections on the surface of metal gears, resulting in a low success rate for feature recognition of gears and mounting holes. Under extreme conditions such as sudden changes in lighting and oil contamination of workpieces, the recognition success rate further decreases, making it very easy for recognition failures to occur in mass production, which in turn leads to assembly interruptions.

[0005] Secondly, in terms of motion control, existing solutions generally adopt uniform motion trajectory planning. In high-speed mode, workpiece collisions and positioning errors are prone to occur, while in low-speed mode, the assembly efficiency is extremely low. It is always impossible to meet the cycle requirements of industrial production and micron-level assembly accuracy, resulting in long assembly time and insufficient positioning accuracy.

[0006] Secondly, at the assembly process level, existing solutions typically require taking photos to obtain the coordinates of the mounting holes first, and then taking a second photo after the robot moves into place to obtain the gear rotation angle and make adjustments. Single-hole assembly requires at least two visual feedbacks, which is cumbersome and results in poor assembly smoothness, making it unsuitable for mass production cycles.

[0007] Furthermore, at the system adaptation level, the existing hand-eye calibration and coordinate transformation schemes are relatively fixed and lack an adaptive correction mechanism. They have poor adaptability to gears with different modules, different outer diameters, and mounting plates with different numbers of holes. When changing products, reprogramming and complete calibration are required, which takes a long time. At the same time, the repeatability of coordinate transformation is low, making it difficult to ensure the consistency of batch assembly.

[0008] Therefore, how to systematically solve the full-dimensional defects of existing gear assembly technology in terms of visual recognition, motion control, assembly process and system adaptation, and achieve the synergistic improvement of precision, efficiency, stability and flexibility of gear precision assembly has become an urgent technical problem to be solved in this field. Summary of the Invention

[0009] The present invention aims to provide a vision servo-based gear precision assembly system and assembly method to systematically solve the defects of existing technologies in vision recognition, motion control, assembly process and system adaptation, and achieve a synergistic improvement in the precision, efficiency, stability and flexibility of gear precision assembly.

[0010] This invention provides a vision servo-based gear precision assembly system, comprising:

[0011] The visual perception layer includes a global shutter monocular industrial camera and a ring shadowless light source, used to acquire images of gears and mounting holes, and deploys a multi-feature fusion anti-interference recognition algorithm to identify the features of gears and mounting holes and solve pose data.

[0012] The decision planning layer is communicatively connected to the visual perception layer. It is used to receive the pose data and perform adaptive hand-eye calibration and coordinate transformation, segmented visual servo trajectory planning, and high-frequency closed-loop visual servo control.

[0013] An execution control layer, comprising a robot and an end effector, is communicatively connected to the decision planning layer and is used to execute gear gripping, posture adjustment, precision meshing, and pressing actions according to the control instructions of the decision planning layer.

[0014] The visual perception layer, decision planning layer, and execution control layer achieve real-time data interaction through industrial Ethernet.

[0015] Furthermore, the multi-feature fusion anti-interference recognition algorithm includes:

[0016] For the mounting holes, the process of "image grayscale conversion → Gaussian filtering for noise reduction → Canny edge detection → Hough circle transform contour extraction" is adopted to fit the output pixel coordinates of the center of the mounting hole, and a first grayscale difference threshold between the hole and the background is set to filter out invalid contours.

[0017] For gears, the effective area of ​​the gear is extracted by threshold segmentation, the coordinates of the gear center pixel are determined by minimum circumcircle fitting, and the gear rotation angle is output by combining tooth pitch measurement and tooth tip / tooth groove feature extraction. A second gray level difference threshold between tooth tip and tooth groove is set to avoid angle calculation deviation. The first gray level difference threshold range is 45-55, and the second gray level difference threshold range is 25-35.

[0018] Furthermore, the segmented visual servo trajectory planning includes a three-segment motion trajectory:

[0019] Rapid traverse segment: moves at the first speed of motion without visual feedback, used for long-distance movement of the gear from the gripping position to the first position above the assembly station;

[0020] Transition section: Moves at a second motion speed and provides intermittent visual feedback at a first feedback frequency for movement from the first position to the second position above the assembly station, and simultaneously completes gear posture pre-adjustment and position coarse calibration in this section;

[0021] Precision assembly section: moves at a third motion speed and provides real-time visual feedback at a second feedback frequency higher than the first feedback frequency, used for the precision assembly process from the second position to the press-fitting position, completing the coaxiality calibration of the gear and the mounting hole, the meshing fit of the gear teeth and the precision press-fitting.

[0022] Wherein, the first movement speed is greater than the second movement speed, the second movement speed is greater than the third movement speed; and the first position is higher than the second position.

[0023] Furthermore, the first movement speed is 45-55 mm / s, the second movement speed is 18-22 mm / s, and the third movement speed is 4-6 mm / s; the first feedback frequency is once every 100 ms, and the second feedback frequency is once every 10 ms; the first position is 50 mm above the assembly station, and the second position is 10 mm above the assembly station.

[0024] Furthermore, the adaptive hand-eye calibration and coordinate transformation specifically include:

[0025] A checkerboard calibration board was used for basic hand-eye calibration to establish a basic transformation matrix between the camera pixel coordinate system and the robot base coordinate system, with a calibration error ≤0.05mm;

[0026] Based on the principle of perspective transformation, two-dimensional pixel coordinates are converted into three-dimensional world coordinates in the robot's base coordinate system;

[0027] Import the prior information of hole center distance from the CAD model of the mounting plate, and use the least squares method to fit and correct the coordinate transformation data so that the coordinate transformation repeatability is ≤0.03mm;

[0028] When changing product models, there is no need for complete recalibration; only the corresponding parameters need to be modified.

[0029] Furthermore, the high-frequency closed-loop visual servo control includes:

[0030] In the precision assembly section, a 10ms-level high-frequency closed-loop vision servo control loop is constructed. The vision system acquires an image and calculates coordinate data every 10ms, and sends the positional deviation between the actual coordinates and the theoretical coordinates to the robot controller in real time.

[0031] The robot controller dynamically corrects the robot's motion path through a PID algorithm, achieving real-time coaxiality calibration and dynamic adaptation of meshing state.

[0032] The proportional coefficient P of the PID algorithm ranges from 0.75 to 0.85, the integral coefficient I ranges from 0.08 to 0.12, and the derivative coefficient D ranges from 0.04 to 0.06.

[0033] This invention also provides a method for precision gear assembly based on vision servoing, comprising the following steps:

[0034] Step S1: Fix the global shutter monocular industrial camera directly above the assembly station, and arrange the ring shadowless light source around the camera to complete the camera parameter configuration and distortion correction;

[0035] Step S2: Establish the basic transformation matrix between the camera pixel coordinate system and the robot base coordinate system using a checkerboard calibration board, and combine the prior information of the hole center distance of the mounting plate CAD model with the least squares method to fit and correct the coordinate transformation data.

[0036] Step S3: Using a dual-fusion recognition scheme of edge gradient features and grayscale histogram features, the first grayscale difference threshold between the mounting hole and the background and the second grayscale difference threshold between the gear tooth tip and the tooth groove are set respectively, and the center coordinates of the mounting hole and the gear rotation angle are identified and output.

[0037] Step S4: Plan a three-segment motion trajectory that precisely matches the speed level and visual feedback level, including a fast movement segment, a transition segment, and a precise assembly segment. The robot moves according to the planned trajectory.

[0038] Step S5: Simultaneously calculate the center coordinates of the mounting hole and the gear rotation angle during the first photo capture. During the rapid movement and transition phases, the robot pre-adjusts the gear posture according to the gear rotation angle to make the gear tooth groove match the spline in the mounting hole. After reaching the end position of the transition phase above the assembly station, it enters the precision assembly phase.

[0039] Step S6: In the precision assembly section, images are acquired in real time at a preset high-frequency cycle and coordinates are calculated. The robot's motion path is dynamically corrected through a PID algorithm until the assembly is in place.

[0040] Further, the three-segment motion trajectory described in step S4 is as follows: the rapid movement segment moves at a first motion speed without visual feedback, used for movement from the gripping position to a first position above the assembly station; the transition segment moves at a second motion speed with intermittent visual feedback at a first feedback frequency, used for movement from the first position to a second position above the assembly station; the precision assembly segment moves at a third motion speed with real-time visual feedback at a second feedback frequency higher than the first feedback frequency, used for movement from the second position to the press-fitting position; wherein, the first motion speed is greater than the second motion speed, which is greater than the third motion speed, and the first position is higher than the second position.

[0041] Furthermore, the first movement speed is 45-55 mm / s, the second movement speed is 18-22 mm / s, and the third movement speed is 4-6 mm / s; the first feedback frequency is once every 100 ms, and the second feedback frequency is once every 10 ms; the first position is 50 mm above the assembly station, and the second position is 10 mm above the assembly station.

[0042] Further, in step S2, the calibration error of the basic transformation matrix is ​​≤0.05mm, the coordinate transformation repeatability is ≤0.03mm, and the product changeover time is ≤30 minutes; in step S3, the first grayscale difference threshold range is 45-55, and the second grayscale difference threshold range is 25-35; in step S6, the high-frequency period is once every 10ms, the proportional coefficient P of the PID algorithm ranges from 0.75 to 0.85, the integral coefficient I ranges from 0.08 to 0.12, and the derivative coefficient D ranges from 0.04 to 0.06.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] (1) This invention adopts an anti-interference recognition scheme that combines edge gradient features and grayscale histogram features, and specifically sets grayscale difference thresholds for mounting holes and gear tooth tips / grooves, effectively suppressing interference from sudden changes in ambient light and reflections on the surface of metal gears. Under normal working conditions, the feature recognition success rate of gears and mounting holes is increased to over 98%, and under extreme working conditions, it still maintains a recognition success rate of over 96%, solving the problem that traditional single-feature recognition schemes are prone to failure under complex working conditions;

[0045] (2) This invention adopts a three-stage trajectory planning strategy that precisely matches speed grading and visual feedback grading. The fast movement stage achieves long-distance efficient movement, the transition stage completes attitude pre-adjustment and position coarse calibration, and the precision assembly stage achieves micron-level precision pressing with high-frequency visual feedback. Compared with the traditional uniform speed trajectory scheme, the total time for 3-hole sequential assembly is controlled within 3 minutes, and the assembly positioning accuracy reaches 0.02mm, realizing a synergistic improvement in production efficiency and assembly accuracy;

[0046] (3) This invention uses a pose-integrated synchronous calculation method to simultaneously acquire the center coordinates of the mounting hole and the gear rotation angle during a single photo capture. The robot pre-adjusts the gear posture during movement to match the phase of the tooth groove with the spline in the mounting hole. Upon reaching the assembly station, it directly completes the meshing and pressing. This eliminates the redundant steps of secondary photo capture in the traditional solution, reduces the single-hole assembly process steps by 40%, and significantly improves the smoothness of assembly.

[0047] (4) This invention adopts an adaptive hand-eye calibration and coordinate transformation method based on prior information of the hole center distance in the CAD model, and uses the least squares method to fit and correct the coordinate transformation matrix. The calibration error is controlled within 0.05mm, the coordinate transformation repeatability reaches 0.03mm, and there is no need to recalibrate completely when changing products. Only the corresponding parameters need to be modified. The changeover time is shortened from 4 hours to less than 30 minutes. It can flexibly adapt to gears with different modules, different outer diameters and different numbers of holes on the mounting plate.

[0048] (5) This invention constructs a high-frequency closed-loop visual servo control loop in the precision assembly section, and dynamically corrects the robot's motion path through the PID algorithm to achieve real-time coaxiality calibration and dynamic adaptation of meshing state, effectively avoiding workpiece jamming and rigid collision. In 100 batch assembly tests, the feature recognition success rate reached 100%, the assembly qualification rate reached 100%, and the batch assembly consistency was significantly better than the existing solutions;

[0049] (6) Through parallel comparison and verification with existing technologies, the present invention reduces the assembly time of 3 holes by 40% compared with the traditional solution, improves the assembly positioning accuracy by 75%, improves the coordinate transformation repeatability accuracy by 62.5%, and reduces the product changeover time by 87.5%. At the same time, it still maintains a high recognition success rate and assembly qualification rate under extreme working conditions such as sudden changes in ambient light, horizontal displacement of the mounting plate, and oil stains and scratches on the gear surface, and the system has strong robustness.

[0050] In summary, this invention makes systematic innovations from multiple dimensions, including visual recognition, motion control, assembly process, system adaptation, and closed-loop control. The various technical features work together and support each other, achieving a synergistic improvement in the precision, efficiency, stability, and flexibility of gear precision assembly. It can be widely used in precision assembly fields such as aerospace, automobile manufacturing, and engineering machinery. Attached Figure Description

[0051] Figure 1 This is a block diagram of the three-layer collaborative architecture of the gear precision assembly system of the present invention;

[0052] Figure 2 This is a flowchart of the multi-feature fusion visual recognition algorithm of the present invention. The left side shows the entire process of mounting hole recognition, and the right side shows the entire process of gear recognition, which is used to illustrate the core logic of dual feature fusion and the grayscale threshold filtering process.

[0053] Figure 3 This is a schematic diagram of the adaptive hand-eye calibration and coordinate transformation principle of the present invention, showing the transformation process from the chessboard calibration board and pixel coordinate system to the robot base coordinate system, as well as the least squares fitting correction logic based on the CAD model.

[0054] Figure 4 This is a flowchart of the entire process of segmented trajectory planning and assembly of the present invention, showing the speed and feedback parameters of the three-segment trajectory, as well as the entire process steps from gear gripping, posture pre-adjustment, integrated meshing assembly to assembly reset, and simultaneously showing the feedback logic of 10ms-level high-frequency closed-loop control. Detailed Implementation

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1

[0057] This embodiment provides a vision servo-based gear precision assembly system, whose three-layer collaborative architecture is as follows: Figure 1 As shown.

[0058] The visual perception layer includes a global shutter monocular industrial camera and a ring-shaped shadowless light source. The camera is fixed directly above the assembly station, and the ring-shaped shadowless light source is arranged around the camera to suppress glare interference from the surface of the metal gears. In this embodiment, the camera is a Hikvision MV-CA013-20GC global shutter camera with a resolution of 1920×1080, a frame rate of 30fps, and an exposure time of 500μs. The camera's parameters are configured and lens distortion is corrected using MVS V3.0 software to ensure that the workpiece image sharpness is not less than 90%.

[0059] The decision-making and planning layer includes an industrial control computer and a vision processing platform and robot control software deployed on it. The vision processing platform uses Vision Master V4.2, and the robot programming software uses AUBO PE V4.5. The decision-making and planning layer receives pose data from the vision perception layer and performs adaptive hand-eye calibration and coordinate transformation, segmented visual servo trajectory planning, integrated pose synchronous calculation, and high-frequency closed-loop visual servo control.

[0060] The execution control layer includes a six-DOF collaborative robot, a pneumatic gripper end effector, a positioning fixture unit, and a motion control drive unit. In this embodiment, the robot is an AUBO-i5 six-DOF collaborative robot with a repeatability of ±0.02mm; the end effector is a pneumatic parallel gripper; and the positioning fixture unit has a positioning accuracy of ≤0.02mm. Real-time data exchange between the layers is achieved via industrial Ethernet, with a transmission latency of ≤5ms.

[0061] Example 2

[0062] This embodiment uses a gear with a module of 2mm, 30 teeth, and an outer diameter of 64mm, and a mounting plate with 3 holes as an example to describe in detail a gear precision assembly method based on vision servoing according to the present invention.

[0063] The specific implementation steps of this embodiment are as follows: hardware setup and tooling debugging, camera parameter debugging, adaptive hand-eye calibration and coordinate transformation, deployment of multi-feature fusion recognition algorithm, segmented trajectory planning and program writing, pose-integrated meshing optimization configuration, and full-process joint debugging and batch verification. The software platform adopts the Windows 10 64-bit operating system, the camera debugging software is MVS V3.0, the vision processing platform is Vision Master V4.2, and the robot programming software is AUBO PE V4.5.

[0064] Step S1: Vision System Hardware Setup and Parameter Debugging

[0065] A global shutter monocular industrial camera was fixed directly above the assembly station, with a ring-shaped shadowless light source surrounding the camera. The focal length was adjusted to ensure clear imaging of the workpiece. The camera resolution was set to 1920×1080, the frame rate to 30fps, and the exposure time to 500μs using the MVS V3.0 camera debugging software. Lens distortion correction was then performed to ensure that the workpiece image sharpness was ≥90%.

[0066] Step S2: Adaptive hand-eye calibration and coordinate transformation

[0067] like Figure 3As shown, a 10mm×10mm checkerboard calibration plate was used, and nine sets of calibration plate images in different poses were acquired. A basic transformation matrix between the camera pixel coordinate system and the robot base coordinate system was established by fitting, with a calibration error ≤0.05mm. Based on the principle of perspective transformation, the two-dimensional pixel coordinates (u,v) were converted into three-dimensional world coordinates (X,Y,Z) in the robot base coordinate system. At the same time, the prior information of the hole center distances from the mounting plate CAD model was imported (e.g., center distance L1 of hole 3 > L2 of hole 4 > L3 of hole 5). The coordinate transformation data was corrected by fitting using the least squares method to eliminate coordinate deviations caused by lens distortion and tooling installation errors. The final coordinate transformation repeatability was ≤0.03mm. When changing products, only the gear module, number of teeth, outer diameter, tooth width parameters, and the number of holes, hole center distances, and hole diameter tolerance parameters of the mounting plate need to be modified. There is no need to re-perform the complete checkerboard calibration, and the changeover time is ≤30 minutes.

[0068] Step S3: Multi-feature fusion recognition and pose calculation

[0069] like Figure 2 As shown, a multi-feature fusion anti-interference recognition algorithm is deployed. Figure 2 The left side shows the mounting hole recognition process: the process of "image grayscale conversion → 5×5 Gaussian filter for noise reduction → Canny edge detection → Hough circle transform contour extraction" is adopted to fit the output pixel coordinates of the center of the mounting hole, and the first grayscale difference threshold between the hole and the background is set to 50 (range 45-55) to filter out invalid contours caused by stray light interference. Figure 2 The right side illustrates the gear recognition process: The effective area of ​​the gear is extracted through threshold segmentation; the coordinates of the gear's center pixel are determined by minimum circumcircle fitting; and the gear rotation angle is output by combining tooth pitch measurement and tooth tip / groove feature extraction. A second grayscale difference threshold of 30 (range 25-35) is set between the tooth tip and groove to avoid angle calculation errors caused by metal reflections. After recognition, the pose data is transmitted in real-time to the decision-making and planning layer via industrial Ethernet, with a transmission delay ≤5ms.

[0070] Steps S4 to S6: Segmented trajectory planning and assembly process

[0071] like Figure 4 As shown, the segmented trajectory planning and assembly process of the present invention includes the following three stages, and the feedback logic of the 10ms-level high-frequency closed-loop control is also shown in the figure.

[0072] Step S4: Segmented visual servo trajectory planning

[0073] A three-stage motion trajectory with precise matching of planned speed levels and visual feedback levels:

[0074] Rapid traverse section: Movement speed 50mm / s (range 45-55mm / s), no visual feedback, used for long-distance movement of gears from the gripping position to 50mm above the assembly station.

[0075] Transition section: Movement speed 20mm / s (range 18-22mm / s), intermittent visual feedback every 100ms, used for movement from 50mm above the assembly station to 10mm above the assembly station. In this section, gear posture pre-adjustment and position coarse calibration are completed simultaneously.

[0076] Precision assembly section: movement speed 5mm / s (range 4-6mm / s), real-time visual feedback every 10ms, used for the precision assembly process from 10mm above the assembly station to the press-fitting position, to complete the coaxiality calibration of gears and mounting holes, tooth meshing fit and precision press-fitting.

[0077] The robot moves along the planned trajectory described above, where the first movement speed (50mm / s) > the second movement speed (20mm / s) > the third movement speed (5mm / s), and the first position (50mm) is higher than the second position (10mm).

[0078] Step S5: Synchronous calculation of pose and meshing assembly

[0079] The synchronous calculation module of the vision processing platform executes the mounting hole recognition algorithm and the gear recognition algorithm simultaneously within a single image acquisition cycle and within the same image frame, synchronously calculating and outputting the coordinates of the mounting hole center and the gear rotation angle. Specifically, in a captured image frame, the Hough circle transform is run simultaneously to obtain the coordinates of the mounting hole center, and the angle between the centroid of the gear profile and the fitted center of the addendum circle is calculated. Combined with tooth groove feature point matching, the current rotation angle of the gear is calculated.

[0080] During the rapid movement and transition phases, the robot's rotational attitude is pre-adjusted synchronously by the robot controller based on the gear rotation angle calculated in real time, ensuring that the gear tooth grooves are perfectly matched with the splines in the mounting holes. After reaching the end position of the transition phase above the assembly station (i.e., the second position, 10mm away from the station), the robot directly enters the precision assembly phase without the need for secondary photography or attitude adjustment.

[0081] Step S6: Real-time correction of high-frequency closed-loop visual servo

[0082] In the precision assembly section, images are acquired in real time with a high-frequency period of 10ms, and coordinate data is calculated. The positional deviation between the actual and theoretical coordinates is sent to the robot controller in real time. The robot controller dynamically corrects the robot's motion path using a PID algorithm, where the proportional coefficient P = 0.8 (range 0.75-0.85), the integral coefficient I = 0.1 (range 0.08-0.12), and the derivative coefficient D = 0.05 (range 0.04-0.06). Figure 4 As shown, this 10ms-level high-frequency closed-loop control loop continuously provides feedback correction, achieving real-time calibration of the coaxiality of the gear and mounting hole and dynamic adaptation of the meshing state until the gear is pressed into place, avoiding workpiece jamming and rigid collisions. After assembly, the robot resets and enters the next assembly cycle.

[0083] Using the above system and method, continuous assembly tests were conducted on 100 sets of gears and mounting plates. Test environment: standard workshop lighting (500 lux), gear surfaces without special treatment. Test results: 100% feature recognition success rate, total assembly time for 3-hole sequential assembly ≤ 3 minutes (average 2 minutes 45 seconds), assembly positioning accuracy ≤ 0.02 mm, and assembly pass rate 100%.

[0084] Even under extreme conditions (ambient light intensity changes abruptly from 500 lux to 5000 lux, mounting plate horizontal offset ±2 mm, and oil stains and scratches on gear surfaces), the feature recognition success rate remains ≥96%, and the assembly qualification rate is 100%.

[0085] The workpiece to be assembled was replaced with a gear with a module of 2.5mm and an outer diameter of 80mm, along with a 5-hole mounting plate. The operator selects the corresponding gear and mounting plate model in the Vision Master V4.2 interface. The system automatically retrieves pre-stored parameters such as gear module, number of teeth, outer diameter, tooth width, and the number of holes, hole center distance, and hole diameter tolerance parameters for the mounting plate, eliminating the need to re-execute the checkerboard calibration process. The changeover takes approximately 25 minutes, significantly less than the 4 hours required by existing solutions.

[0086] The solution of this invention was compared in parallel with the traditional single-feature recognition + uniform trajectory + secondary image capture solution, and the results are as follows:

[0087]

[0088] This invention is compatible with gears with a module of 1.5-3.0mm and an outer diameter of 50-150mm, as well as mounting plates with 2-5 holes. By modifying the corresponding parameters in the vision platform, flexible assembly of products of different specifications can be achieved without recalibration and programming, and it can be widely used in precision assembly fields such as aerospace, automobile manufacturing, and engineering machinery.

[0089] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the inventive concept of the present invention, and all such modifications or additions should fall within the protection scope of the present invention.

Claims

1. A visual servoing based precision gear assembly system, characterized in that, include: The visual perception layer includes a global shutter monocular industrial camera and a ring shadowless light source, used to acquire images of gears and mounting holes, and deploys a multi-feature fusion anti-interference recognition algorithm to identify the features of gears and mounting holes and solve pose data. The multi-feature fusion anti-interference recognition algorithm includes: For the mounting holes, the process of "image grayscale conversion → Gaussian filtering for noise reduction → Canny edge detection → Hough circle transform contour extraction" is adopted to fit the output pixel coordinates of the center of the mounting hole, and a first grayscale difference threshold between the hole and the background is set to filter out invalid contours. For gears, the effective area of ​​the gear is extracted by threshold segmentation, the coordinates of the gear center pixel are determined by minimum circumcircle fitting, and the gear rotation angle is output by combining tooth pitch measurement and tooth tip / tooth groove feature extraction. A second gray level difference threshold between tooth tip and tooth groove is set to avoid angle calculation deviation. The first grayscale difference threshold range is 45-55, and the second grayscale difference threshold range is 25-35; The decision planning layer is communicatively connected to the visual perception layer. It is used to receive the pose data and perform adaptive hand-eye calibration and coordinate transformation, segmented visual servo trajectory planning, and high-frequency closed-loop visual servo control. The adaptive hand-eye calibration and coordinate transformation imports the prior information of the hole center distance of the CAD model of the mounting plate and corrects the coordinate transformation data by fitting the least squares method. The segmented visual servo trajectory planning includes a three-segment motion trajectory: Rapid traverse segment: moves at the first speed of motion without visual feedback, used for long-distance movement of the gear from the gripping position to the first position above the assembly station; Transition section: Moves at a second motion speed and provides intermittent visual feedback at a first feedback frequency for movement from the first position to the second position above the assembly station, and simultaneously completes gear posture pre-adjustment and position coarse calibration in this section; Precision assembly section: moves at a third motion speed and provides real-time visual feedback at a second feedback frequency higher than the first feedback frequency, used for the precision assembly process from the second position to the press-fitting position, completing the coaxiality calibration of the gear and the mounting hole, the meshing fit of the gear teeth and the precision press-fitting. Wherein, the first movement speed is greater than the second movement speed, the second movement speed is greater than the third movement speed; the first position is higher than the second position; The high-frequency closed-loop visual servo control includes: constructing a 10ms-level high-frequency closed-loop visual servo control loop in the precision assembly section; the vision system acquires an image and calculates coordinate data every 10ms; and sends the position deviation between the actual coordinates and the theoretical coordinates to the robot controller in real time; the robot controller dynamically corrects the robot's motion path through a PID algorithm to achieve real-time coaxiality calibration and dynamic adaptation of the meshing state. An execution control layer, comprising a robot and an end effector, is communicatively connected to the decision planning layer and is used to execute gear gripping, posture adjustment, precision meshing, and pressing actions according to the control instructions of the decision planning layer. The visual perception layer, decision planning layer, and execution control layer achieve real-time data interaction through industrial Ethernet.

2. The vision servo-based gear precision assembly system according to claim 1, characterized in that, The first movement speed is 45-55 mm / s, the second movement speed is 18-22 mm / s, and the third movement speed is 4-6 mm / s; the first feedback frequency is once every 100 ms, and the second feedback frequency is once every 10 ms; the first position is 50 mm above the assembly station, and the second position is 10 mm above the assembly station.

3. The vision servo-based gear precision assembly system according to claim 1, characterized in that, The adaptive hand-eye calibration and coordinate transformation specifically include: A checkerboard calibration board was used for basic hand-eye calibration to establish a basic transformation matrix between the camera pixel coordinate system and the robot base coordinate system, with a calibration error ≤0.05mm; Based on the principle of perspective transformation, two-dimensional pixel coordinates are converted into three-dimensional world coordinates in the robot's base coordinate system; Import the prior information of hole center distance from the CAD model of the mounting plate, and use the least squares method to fit and correct the coordinate transformation data so that the coordinate transformation repeatability is ≤0.03mm; When changing product models, there is no need to recalibrate completely; only the corresponding parameters need to be modified. These corresponding parameters include the gear's module, number of teeth, outer diameter, and tooth width, as well as the mounting plate's number of holes, hole center distance, and hole diameter tolerance.

4. The vision servo-based gear precision assembly system according to claim 1, characterized in that, In the high-frequency closed-loop visual servo control, the proportional coefficient P of the PID algorithm ranges from 0.75 to 0.85, the integral coefficient I ranges from 0.08 to 0.12, and the derivative coefficient D ranges from 0.04 to 0.

06.

5. A method for precision gear assembly based on vision servoing, characterized in that, The method is implemented using a vision-servo-based gear precision assembly system as described in any one of claims 1 to 4, and includes the following steps: Step S1: Fix the global shutter monocular industrial camera directly above the assembly station, and arrange the ring shadowless light source around the camera to complete the camera parameter configuration and distortion correction; Step S2: Establish the basic transformation matrix between the camera pixel coordinate system and the robot base coordinate system using a checkerboard calibration board, and combine the prior information of the hole center distance of the mounting plate CAD model with the least squares method to fit and correct the coordinate transformation data. Step S3: Using a dual-fusion recognition scheme of edge gradient features and grayscale histogram features, the first grayscale difference threshold between the mounting hole and the background and the second grayscale difference threshold between the gear tooth tip and the tooth groove are set respectively, and the center coordinates of the mounting hole and the gear rotation angle are identified and output. Step S4: Plan a three-segment motion trajectory that precisely matches the speed level and visual feedback level, including a fast movement segment, a transition segment, and a precise assembly segment. The robot moves according to the planned trajectory. Step S5: Through the synchronous calculation module of the vision processing platform, the mounting hole recognition algorithm and the gear recognition algorithm are executed synchronously within a single image acquisition cycle and in the same frame image, and the center coordinates of the mounting hole and the rotation angle of the gear are calculated and output synchronously. During the movement of the robot in the rapid movement segment and the transition segment, the rotation attitude of the end effector is pre-adjusted synchronously by the robot controller according to the gear rotation angle obtained by synchronous calculation, so that the phase of the gear tooth groove and the spline in the mounting hole is completely matched. After reaching the end position of the transition segment above the assembly station, it directly enters the precision assembly segment. Step S6: In the precision assembly section, images are acquired in real time with a preset high-frequency period of 10ms and coordinate data is calculated. The positional deviation between the actual coordinates and the theoretical coordinates is compared. The robot's motion path is dynamically corrected through the PID algorithm to complete the gear coaxiality calibration and meshing press-fit until the press-fit is in place.

6. The gear precision assembly method based on vision servoing according to claim 5, characterized in that, The three-segment motion trajectory described in step S4 is as follows: the rapid movement segment moves at a first motion speed without visual feedback and is used for movement from the gripping position to a first position above the assembly station; the transition segment moves at a second motion speed and provides intermittent visual feedback at a first feedback frequency and is used for movement from the first position to a second position above the assembly station. The precision assembly section moves at a third motion speed and provides real-time visual feedback at a second feedback frequency higher than the first feedback frequency, for moving from the second position to the press-fitting position; wherein, the first motion speed is greater than the second motion speed, which is greater than the third motion speed, and the first position is higher than the second position.

7. The gear precision assembly method based on vision servoing according to claim 6, characterized in that, The first movement speed is 45-55 mm / s, the second movement speed is 18-22 mm / s, and the third movement speed is 4-6 mm / s; the first feedback frequency is once every 100 ms, and the second feedback frequency is once every 10 ms; the first position is 50 mm above the assembly station, and the second position is 10 mm above the assembly station.

8. The gear precision assembly method based on vision servoing according to claim 5, characterized in that, In step S2, the calibration error of the basic transformation matrix is ​​≤0.05mm, the coordinate transformation repeatability is ≤0.03mm, and the product changeover time is ≤30 minutes; in step S3, the first grayscale difference threshold range is 45-55, and the second grayscale difference threshold range is 25-35; in step S6, the high-frequency period is once every 10ms, the proportional coefficient P of the PID algorithm ranges from 0.75 to 0.85, the integral coefficient I ranges from 0.08 to 0.12, and the derivative coefficient D ranges from 0.04 to 0.06.

Images