Double-station solder ball welding machine correction system based on machine vision and multi-axis cooperation

Abstract

The invention discloses a double-station solder ball welding machine correction system based on machine vision and multi-axis cooperation, and belongs to the field of welding equipment correction. A double-station solder ball welding machine correction system based on machine vision and multi-axis cooperation comprises an image acquisition unit, a visual processing unit, a multi-axis control unit and a cooperative scheduling unit. The problems that in existing double-station welding equipment, due to the fact that a single-station vision scheme is simply copied, the equipment takt is slow, the productivity is limited, and comprehensive compensation for dynamic errors and optical environment differences under multi-station collaborative operation is lacked are solved. Compared with the prior art, the double-station equipment has the advantages that overlapped execution of visual correction and welding operation is achieved, and feed-forward and online compensation mechanisms for dynamic errors such as thermal deformation and vibration are integrated, so that high-precision positioning is guaranteed, and the production efficiency and the operation reliability of the double-station equipment are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of welding equipment straightening technology, specifically to a dual-station solder ball welding machine straightening system based on machine vision and multi-axis collaboration. Background Technology

[0002] In the fields of microelectronics packaging and advanced manufacturing, high-density, miniaturized solder ball soldering (such as BGA and CSP packages) is a key process to ensure the reliability of circuit interconnects. Traditional single-station solder ball soldering equipment usually relies on preset programs for soldering, which is difficult to adapt to microscopic shifts in pad position caused by PCB board deformation, cumulative fixture positioning errors, thermal drift, etc., easily leading to solder ball misalignment, bridging, or cold solder joints, seriously affecting yield and product reliability. Introducing machine vision for position correction has become the mainstream direction for improving accuracy. Existing technologies mostly adopt a serial process of "photographing-calculating-compensating-soldering". However, in dual-station equipment, simply replicating the single-station vision solution will result in long visual processing and mechanical motion waiting times, slow equipment cycle time, and limited production capacity. In addition, when dual-stations operate alternately, dynamic factors such as mechanical vibration and thermal disturbance will introduce new error sources, and existing systems often lack a comprehensive compensation mechanism for multi-station collaboration and dynamic errors. Therefore, it does not meet the current needs. To address this, we propose a dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration. Summary of the Invention

[0003] The purpose of this invention is to provide a correction system for a dual-station solder ball welding machine based on machine vision and multi-axis collaboration. By integrating parallel image processing, multi-axis collaborative motion planning, and intelligent collaborative scheduling, it achieves efficient overlapping execution of dual-station visual correction and welding operations in time. This not only overcomes the problem of slow equipment cycle time caused by traditional serial processes and significantly improves production capacity, but also effectively addresses the complex errors introduced by dynamic factors such as alternating dual-station operations, mechanical vibration, and thermal drift by introducing global pre-compensation, dynamic error identification and compensation, and deterministic and adaptive collaborative scheduling mechanisms, thus solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration, comprising: The image acquisition unit is configured to simultaneously acquire images of the workpiece pads on the first and second workstations. The vision processing unit is configured to receive and process the pad images of the first station and the second station in parallel, for identifying the actual position of the solder joints at each station and calculating the offset vector of the actual position of the solder joints at each station relative to the preset calibration position. The multi-axis control unit is configured to generate coordinated motion commands to drive the welding head for position compensation based on the received offset vector; The collaborative scheduling unit is configured to alternately trigger the image acquisition and processing flow of the first and second workstations according to a preset rhythm, and to schedule the multi-axis control unit to drive the welding head to perform position compensation and welding operations corresponding to the offset vectors on the first and second workstations in sequence, so as to realize the parallel and alternating operation of visual correction and welding operations of the two workstations.

[0005] Furthermore, the image acquisition unit includes: The synchronous control module receives unified acquisition instructions from the collaborative scheduling unit, generates two synchronous hardware trigger signals, and controls the industrial cameras set at the first and second workstations to acquire images respectively. The lighting adjustment module is configured to independently and dynamically adjust the intensity, spectrum, and incident angle of the light source set at the first and second workstations based on online monitoring of the reflectivity of the fixtures at different workstations and changes in ambient light.

[0006] Furthermore, the visual processing unit includes: The image preprocessing module is configured to use a parallel computing architecture to perform synchronous processing on the workpiece pad images of the two workstations, including grayscale processing, noise reduction processing and image enhancement processing. The image enhancement processing includes contrast stretching and histogram equalization. The feature recognition module is configured to locate and segment the solder joint region in the preprocessed workpiece solder pad image based on a trained solder joint recognition model. The solder joint recognition model is an image recognition model based on a deep convolutional neural network. The coordinate calculation module is configured to extract the geometric center coordinates of each solder joint area as the actual position, and convert the image pixel coordinates into actual physical coordinates in the machine motion coordinate system according to the preset camera calibration parameters, and then calculate the offset vector.

[0007] Furthermore, after calculating the offset vector, the vision processing unit performs statistical analysis on the offset vectors of the first and second workstations obtained in the continuous production cycle. If an overall offset pattern related to the batch clamping error of the workpiece is identified, global position compensation parameters are generated and sent to the multi-axis control unit for batch pre-compensation of the welding path of subsequent workpieces.

[0008] Furthermore, the multi-axis control unit includes: The motion planning module is configured to plan a smooth, singularity-free movement trajectory of the welding head in three-dimensional space based on the pre-stored weld point sequence coordinates in the welding path process library, the safe height of the welding head approaching and lifting, and the received offset vector, taking into account the acceleration and deceleration capabilities of each motion axis and the maximum jump constraint, and generating corresponding timing commands for the position, velocity, and acceleration of each axis. The position compensation module is configured to decompose the offset vector corresponding to each solder point into independent motion components of the X-axis, Y-axis, Z-axis and R-axis according to the machine kinematics model during the execution of the motion trajectory. The anti-collision monitoring module is configured to predict and prevent possible interference between the welding head and the fixture during the dual-station switching process by monitoring the current and position feedback of each axis motor.

[0009] Furthermore, a spline curve interpolation algorithm is used to plan a smooth, singularity-free trajectory for the welded joint in three-dimensional space, including: After receiving the theoretical coordinates of the weld point sequence and the real-time offset vector, the complete welding path is automatically divided into two types of continuous sub-trajectory segments: the two types of continuous sub-trajectory segments include the welding segment within the workstation and the transfer segment between workstations. The welding segment within the workstation and the transfer segment between workstations included in the three-dimensional spatial motion trajectory of the welding head are decoupled into a planar trajectory and a height trajectory, respectively. The planar and height trajectories of the welding section within the workstation and the transfer section between workstations are smoothed using spline curve interpolation algorithms to obtain the smoothed planar and height trajectories of the welding section within the workstation and the transfer section between workstations. The planar and height trajectories of the welding section within the workstation and the transfer section between workstations after smoothing are integrated to form a smooth, singularity-free movement trajectory of the welding head in three-dimensional space.

[0010] Furthermore, spline curve interpolation algorithms are used to smooth the planar and height trajectories of the welding section within the workstation and the transfer section between workstations, respectively, to obtain the smoothed planar and height trajectories of the welding section within the workstation and the transfer section between workstations, including: Iterate through all the weld points contained in the welding section within the workstation in sequence; The trajectory segments corresponding to the planar trajectories between every two weld points in the welding section within the workstation are smoothed by using quintic spline interpolation, thereby obtaining the smoothed trajectory segments corresponding to the planar trajectories between every two weld points in the welding section within the workstation. The height of the welding head remains constant throughout the trajectory segment corresponding to every two welding points within the welding section at the workstation. Iterate through all the solder joints contained in the transfer section between workstations in sequence; The trajectory segments corresponding to the planar trajectories between every two welding points in the planned obstacle avoidance path within the inter-station transfer section are smoothed by using quintic spline curve interpolation. Cubic spline curve interpolation is used to smooth the trajectory segment corresponding to the height trajectory between every two welding points in the inter-station transfer section, thereby obtaining the smoothed trajectory segment corresponding to the height trajectory between every two welding points in the inter-station transfer section.

[0011] Furthermore, during execution, the motion planning module continuously collects actual motion trajectories, compensation effects, and welding quality data from historical welding cycles. It trains a strategy network using an online reinforcement learning algorithm to iteratively optimize the smoothness of the motion trajectory, velocity curve, and acceleration parameters.

[0012] Furthermore, the position compensation module introduces an error compensation mechanism on the basis of performing feedforward compensation, which is used to identify and compensate online for low-frequency quasi-static errors caused by thermal deformation of the mechanical transmission chain and straightness error of the guide rail, as well as high-frequency dynamic errors introduced by structural vibration caused by alternating motion of the two workstations.

[0013] Furthermore, the coordinated scheduling unit includes: The timing management module is configured to generate and execute a time-triggered deterministic scheduling timing table, and through a hardware clock synchronization mechanism, ensure that the vision processing stage of the first station and the welding stage of the second station overlap in time during the cycle of alternating operations of the two stations, while maintaining phase locking in coordination with the mechanical vibration cycle to form a deterministic assembly line operation cycle. The status monitoring module is configured to periodically collect the trigger count of the image acquisition unit, the feature recognition confidence of the vision processing unit, the actual position and following error of each axis of the multi-axis control unit, and the voltage and current waveform data of the welding power supply. The module compares the above data with the preset process thresholds in real time, and generates a hierarchical alarm information log containing fault codes, timestamps and possible causes when the data exceeds the limit.

[0014] Furthermore, the collaborative scheduling unit also introduces a fault self-recovery mechanism, specifically including: Based on the received hierarchical alarm information, execute a tiered response strategy; For non-critical parameter drift alarms, the process parameters of subsequent cycles are fine-tuned according to the preset compensation rule library without interrupting the current work cycle. For critical motion and process fault alarms, immediately initiate a safety interruption sequence, freeze the motion of each axis, save the field status, and automatically execute the preset recovery process; After the recovery process is completed, based on the processing status and compensation data completed before the interruption, the execution timing and path of the remaining solder joints are recalculated and optimized to achieve seamless production recovery from the breakpoint. In subsequent scheduling, the identified fault modes are automatically avoided and the resource allocation strategy is dynamically adjusted.

[0015] Furthermore, at the beginning of each scheduling cycle of the timing management module, the collaborative scheduling unit sends a synchronization instruction containing a global cycle number and a timestamp to the image acquisition unit, vision processing unit, multi-axis control unit, and welding power supply. It receives process data with timing marks from the above units and constructs a unified data frame corresponding to each workstation and each processing cycle in memory based on the global cycle number and timestamp. Before the status monitoring module performs real-time comparison, it performs logical consistency verification on the related data items in the unified data frame. If the verification fails, the frame data is discarded and a re-sampling mechanism for a specific data source is triggered. At the same time, interpolation prediction is performed based on the previous valid data frame.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention significantly shortens the visual processing cycle and eliminates equipment waiting time caused by serial processes by synchronously acquiring and parallel processing images from two workstations, and combining image preprocessing and compression with edge computing nodes deployed at the front end. At the same time, the system utilizes a collaborative scheduling unit to achieve precise, overlapping, pipeline-style execution of visual correction and welding operations between the two workstations, significantly improving the overall cycle time and production capacity of the equipment. Furthermore, this invention not only achieves real-time position compensation based on single measurements, but also identifies and pre-compensates batch clamping errors through the statistical analysis capability of the offset vector by the visual processing unit. The low-frequency and high-frequency error feedforward-feedback composite compensation mechanism introduced by the multi-axis control unit effectively suppresses errors introduced by mechanical thermal deformation and dynamic vibration. Through the deterministic timing management, status monitoring, and fault self-recovery mechanism of the collaborative scheduling unit, high reliability, high stability, and high yield are ensured for the two workstations under complex alternating operation conditions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of the dual-station solder ball welding machine correction system based on machine vision and multi-axis collaboration of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To address the issues of slow cycle time and limited capacity in existing dual-station welding equipment due to the simple replication of single-station vision solutions, and the lack of comprehensive compensation for dynamic errors and optical environment differences in multi-station collaborative operations, please refer to [link to relevant documentation]. Figure 1 This embodiment provides the following technical solution: A dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration includes: The image acquisition unit is configured to simultaneously acquire images of the workpiece pads on the first and second workstations. The vision processing unit is configured to receive and process the pad images of the first station and the second station in parallel, for identifying the actual position of the solder joints at each station and calculating the offset vector of the actual position of the solder joints at each station relative to the preset calibration position. The multi-axis control unit is configured to generate coordinated motion commands to drive the welding head for position compensation based on the received offset vector; The collaborative scheduling unit is configured to alternately trigger the image acquisition and processing flow of the first and second workstations according to a preset rhythm, and to schedule the multi-axis control unit to drive the welding head to perform position compensation and welding operations corresponding to the offset vectors on the first and second workstations in sequence, so as to realize the parallel and alternating operation of visual correction and welding operations of the two workstations. The image acquisition unit and the vision processing unit are connected by a distributed edge computing node. The node is configured to perform an image format conversion operation locally before the image acquisition unit uploads the acquired workpiece pad image to the vision processing unit. It also uses a lossless compression algorithm based on predictive coding to compress the converted workpiece pad image data before uploading it to the vision processing unit.

[0020] The technical effects of the above solution are as follows: By synchronously acquiring images of the solder pads at both workstations through the image acquisition unit, and combining this with the vision processing unit to process and identify the actual position and offset vector of the solder joints in parallel, the welding head is driven by the multi-axis control unit for precise position compensation. Then, the visual correction and welding process of the two workstations is triggered alternately by the collaborative scheduling unit according to the rhythm, which can realize efficient parallel and alternating operation of the two workstations. At the same time, by introducing distributed edge computing nodes between the image acquisition unit and the vision processing unit, the image format is converted locally and the image is compressed and uploaded using a lossless compression algorithm based on predictive coding, which can reduce the data transmission load and latency, thereby improving the overall response speed and operating efficiency of the system, and finally achieving rapid, accurate collaborative correction and continuous production of the solder ball welding process at both workstations.

[0021] The image acquisition unit includes: The synchronous control module receives unified acquisition instructions from the collaborative scheduling unit, generates two synchronous hardware trigger signals, and controls the industrial cameras set at the first and second workstations to acquire images respectively. The lighting adjustment module is configured to independently and dynamically adjust the intensity, spectrum and incident angle of the light source set at the first and second workstations based on online monitoring of the reflectivity of the fixtures at different workstations and changes in ambient light. This is used to eliminate the inconsistency in imaging quality caused by the inherent differences in the optical environment between the two workstations and to provide the vision processing unit with image input with uniform lighting conditions. The lighting adjustment module monitors the ambient light intensity and spectral distribution of each workstation in real time using light sensors integrated into each workstation. Simultaneously, it analyzes the reflectivity characteristics of the fixture surface using preview images captured by a camera. The monitored and analyzed ambient light and reflectivity data are compared with preset baseline optical parameters to generate independent adjustment commands for each workstation. These commands are sent to the corresponding workstation's light source controller, which independently changes the light source intensity by adjusting the LED current, alters the spectrum by switching LED combinations with different color temperatures or adjusting filters, and changes the spatial orientation of the light source module by controlling a servo motor to change the light incidence angle. After adjustment, the system re-acquires images to evaluate image quality. If the uniformity standard is not met, the system enters the next fine-tuning cycle, dynamically ensuring consistent image input under uniform lighting conditions for both workstations.

[0022] The technical effects of the above solution are as follows: the synchronization control module generates a synchronization hardware trigger signal by receiving a unified instruction, which can ensure the strict time consistency of the images acquired by the dual-station industrial cameras, thus laying the foundation for subsequent parallel processing. The lighting adjustment module can eliminate the problem of inconsistent optical imaging caused by the difference between the reflectivity of the fixture and the ambient light between the two stations by monitoring online and independently and dynamically adjusting the intensity, spectrum and incident angle of the light source of each station. This provides the vision processing unit with high-quality image input with uniform lighting conditions, thereby ensuring the accuracy and reliability of subsequent position recognition and offset calculation.

[0023] The visual processing unit includes: The image preprocessing module is configured to use a parallel computing architecture to perform synchronous processing on the workpiece pad images of the two workstations, including grayscale processing, noise reduction processing and image enhancement processing. The image enhancement processing includes contrast stretching and histogram equalization. The feature recognition module is configured to locate and segment the solder joint region in the preprocessed workpiece solder pad image based on a trained solder joint recognition model. The solder joint recognition model is an image recognition model based on a deep convolutional neural network. The coordinate calculation module is configured to extract the geometric center coordinates of each solder joint area as the actual position, and convert the image pixel coordinates into actual physical coordinates in the machine motion coordinate system according to the preset camera calibration parameters, and then calculate the offset vector. The vision processing unit calculates the offset vector and then performs statistical analysis on the offset vectors of the first and second workstations obtained in the continuous production cycle. If it identifies the overall offset pattern related to the batch clamping error of the workpiece, it generates global position compensation parameters and sends them to the multi-axis control unit for batch pre-compensation of the welding path of subsequent workpieces.

[0024] In this embodiment, the coordinate calculation module receives the set of contour pixel coordinates of each solder joint area output by the feature recognition module; based on the set of contour pixel coordinates, the geometric center pixel coordinates of the solder joint area in the image are obtained by calculating the average X-coordinate and average Y-coordinate of all contour pixels; the lens distortion correction of the geometric center pixel coordinates is performed by calling the pre-calibrated and stored camera intrinsic parameter matrix and distortion coefficients; the distortion-corrected pixel coordinates are transformed to three-dimensional physical coordinates in the robot base coordinate system or machine motion coordinate system using a preset hand-eye calibration transformation matrix, and these coordinates are the actual physical position of the solder joint; the theoretical design coordinates of the corresponding solder pad are read from the system process file, and the calculated actual physical coordinates are vector subtracted from the theoretical design coordinates to obtain the X-direction and Y-direction offset vectors used to guide the motion compensation of the welding head.

[0025] The technical effects of the above solution are as follows: The image preprocessing module performs grayscale conversion, noise reduction, and image enhancement processing including contrast stretching and histogram equalization on the dual-station images simultaneously through a parallel computing architecture, which can improve image quality and consistency, thereby creating conditions for feature recognition. The feature recognition module, with the help of a weld point recognition model trained on a deep convolutional neural network, achieves accurate positioning and segmentation of the weld point region in the preprocessed image, thereby improving the robustness of recognition under complex conditions. The coordinate calculation module accurately calculates the actual physical coordinates and offset vector of the weld point through coordinate transformation and geometric center extraction. Furthermore, the vision processing unit can intelligently identify the overall offset pattern related to the batch clamping error of the workpiece through statistical analysis of the offset vector of the dual-station in the continuous production cycle, and generate global position compensation parameters accordingly, which are sent to the multi-axis control unit to realize batch pre-compensation for the welding path of subsequent workpieces, thereby improving the accuracy of a single operation.

[0026] Multi-axis control unit, including: The motion planning module is configured to plan a smooth, singularity-free movement trajectory of the welding head in three-dimensional space based on the pre-stored weld point sequence coordinates in the welding path process library, the safe height of the welding head approaching and lifting, and the received offset vector, taking into account the acceleration and deceleration capabilities of each motion axis and the maximum jump constraint, and generating corresponding timing commands for the position, velocity, and acceleration of each axis. During execution, the motion planning module continuously collects actual motion trajectories, compensation effects, and welding quality data from historical welding cycles. It trains a strategy network using an online reinforcement learning algorithm to iteratively optimize the smoothness of the motion trajectory, velocity curves, and acceleration parameters. The position compensation module is configured to decompose the offset vector corresponding to each solder point into independent motion components of the X-axis, Y-axis, Z-axis and R-axis according to the machine kinematics model during the execution of the motion trajectory. Among them, the position compensation module introduces an error compensation mechanism on the basis of performing feedforward compensation, which is used to identify and compensate for the low-frequency quasi-static error of the offset vector caused by thermal deformation of mechanical transmission chain and straightness error of guide rail, as well as the high-frequency dynamic error introduced by structural vibration caused by alternating motion of dual workstations. The anti-collision monitoring module is configured to predict and prevent possible interference between the welding head and the fixture during the dual-station switching process by monitoring the current and position feedback of each axis motor.

[0027] The technical effects of the above-mentioned solutions are as follows: The motion planning module, by combining pre-stored welding paths, safety parameters, and offset vectors, can plan a smooth, singularity-free three-dimensional movement trajectory using spline curve interpolation under strict dynamic constraints, thereby ensuring the accuracy and efficiency of the welding process. Furthermore, by integrating online reinforcement learning capabilities and continuously collecting production data and training the strategy network, it achieves dynamic iterative optimization of the smoothness of the motion trajectory and motion parameters, thereby continuously improving welding quality and system adaptability. The position compensation module decomposes the offset vector into independent motion components for each axis and, by integrating feedforward and error compensation mechanisms, can suppress low-frequency errors caused by mechanical thermal deformation and guide rail errors, as well as high-frequency errors caused by structural vibration, thereby improving the overall accuracy and stability of position compensation. The anti-collision monitoring module actively predicts and prevents potential interference risks during dual-station switching by monitoring motor current and position feedback in real time, thereby enhancing the safety and reliability of system operation.

[0028] Specifically, a spline curve interpolation algorithm is used to plan a smooth, singularity-free trajectory for the welded joint in three-dimensional space, including: After receiving the theoretical coordinates of the weld point sequence and the real-time offset vector, the complete welding path is automatically divided into two types of continuous sub-trajectory segments: the two types of continuous sub-trajectory segments include the welding segment within the workstation and the transfer segment between workstations; specifically, the welding segment within the workstation refers to the movement trajectory of the joint moving sequentially to each weld point within the same workstation; the transfer segment between workstations refers to the trajectory of the welding head being raised and quickly moved above the starting point of another workstation after completing all the welding at one workstation. The welding segment within the workstation and the transfer segment between workstations included in the three-dimensional spatial motion trajectory of the welding head are decoupled into a planar trajectory and a height trajectory, respectively. The planar and height trajectories of the welding section within the workstation and the transfer section between workstations are smoothed using spline curve interpolation algorithms to obtain the smoothed planar and height trajectories of the welding section within the workstation and the transfer section between workstations. The planar and height trajectories of the welding section within the workstation and the transfer section between workstations after smoothing are integrated to form a smooth, singularity-free movement trajectory of the welding head in three-dimensional space.

[0029] This embodiment employs a spline curve interpolation algorithm to directly eliminate singularities in the trajectory, ensuring continuous and smooth three-dimensional movement of the welding head. This avoids movement stutters and abrupt changes, guaranteeing the stability of the welding head's operation and preventing mechanical impacts and positioning deviations caused by trajectory singularities. Simultaneously, it automatically divides the welding path into welding segments within each workstation and transfer segments between workstations, achieving precise segmentation planning. This matches the motion logic of both types of trajectory segments with operational requirements, aligning with the precise requirements of welding operations within each workstation and meeting the high-efficiency requirements of rapid transfer between workstations, thus avoiding the homogenization of trajectory planning. The three-dimensional trajectory is decoupled into planar and height trajectories, which are then smoothed separately, reducing the complexity of three-dimensional trajectory planning and enabling independent and precise optimization of each dimension of the trajectory. This ensures positioning accuracy in the planar direction and smoothness of movement in the height direction, improving the overall accuracy of trajectory planning. Meanwhile, by integrating the smoothed trajectories of each dimension and segment, a complete three-dimensional smooth trajectory is formed, ensuring that the movement of the welding head from welding within the workstation to the transfer between workstations is seamless, taking into account both the process accuracy of welding operations and the efficiency of workstation switching. At the same time, the algorithm can adapt to the theoretical coordinates of the weld point and the real-time offset vector, improving the real-time performance and adaptability of trajectory planning, and adapting to the coordinate deviation of actual welding scenarios.

[0030] Specifically, spline curve interpolation algorithms are used to smooth the planar and height trajectories of the welding section within the workstation and the transfer section between workstations, respectively, to obtain the smoothed planar and height trajectories of the welding section within the workstation and the transfer section between workstations, including: Iterate through all the weld points contained in the welding section within the workstation in sequence; The trajectory segments corresponding to the planar trajectories between every two weld points in the welding section within the workstation are smoothed by using quintic spline interpolation, thereby obtaining the smoothed trajectory segments corresponding to the planar trajectories between every two weld points in the welding section within the workstation. The trajectory segment corresponding to the planar trajectory between every two weld points in the smoothed welding section within the workstation is represented by the following planar trajectory function: Lws(τ)=(a 0x a 0y ) T+(a 1x a 1y ) T τ+(a 2x a 2y ) T τ 2 +(a 3x a 3y ) T τ 3 +(a 4x a 4y ) T τ 4 +(a 5x a 5y ) T τ 5 Where Lws(τ) represents the trajectory function corresponding to the trajectory segment of the planar trajectory between every two weld points in the smoothed welding section within the workstation; (a 0x a 0y ), (a 1x a 1y ), (a 2x a 2y ), (a 3x a 3y ), (a 4x a 4y ) and (a 5x a 5y The vector represents the planar trajectory spline coefficient vector of the quintic spline curve. The elements with x and y subscripts are the weighted components of the planar trajectory spline coefficient vector along the x and y axes, respectively. These weighted components are set according to the boundary conditions corresponding to the welding start and end points, velocity, and acceleration. Furthermore, the planar trajectory spline coefficient vector is used to transform discrete weld point coordinate data into a continuous and smooth spatial trajectory; T represents the transpose operation corresponding to the planar trajectory spline coefficient vector; τ represents the time independent variable. The height of the welding head remains constant throughout the trajectory segment corresponding to every two welding points within the welding section at the workstation. Iterate through all the solder joints contained in the transfer section between workstations in sequence; The trajectory segments corresponding to the planar trajectories between every two weld points in the planned obstacle avoidance path within the inter-station transfer section are smoothed by using quintic spline interpolation. The smoothing method is the same as the smoothing method for the planar trajectories between every two weld points in the welding section within the aforementioned workstation. The trajectory segment corresponding to the height trajectory between each two welding points in the inter-station transfer section is smoothed by using cubic spline curve interpolation. The smoothed trajectory segment corresponding to the height trajectory between each two welding points in the inter-station transfer section is obtained. The height trajectory segment between every two welding points in the smoothed inter-station transfer segment is represented by the following height trajectory function: Among them, Z TS (τ) represents the trajectory segment corresponding to the height trajectory between every two welding points in the smoothed inter-station transfer section; Z s Indicates the initial safe height of the transfer section between workstations; Z safe Z represents the absolute safety platform height of the transfer section between workstations. e τ1 represents the endpoint safety height of the transfer section between workstations; τ2 represents the moment when the welding head reaches the absolute safety height and begins to maintain it; τ3 represents the moment when the welding head begins to descend from the platform.

[0031] This embodiment introduces boundary conditions for position, velocity, and acceleration, generating a continuous and smooth trajectory that ensures no sudden velocity changes or acceleration fluctuations as the welding head moves between weld points, avoiding mechanical impact and significantly improving welding positioning accuracy and weld formation quality. Simultaneously, the constant height of the welding head ensures process stability and weld consistency. Furthermore, a differentiated interpolation strategy is employed for the inter-station transfer section. The planar trajectory reuses quintic spline curves to ensure smoothness during rapid movement, while the height trajectory uses cubic spline curves for segmented control. This design satisfies the efficiency requirements of rapid inter-station transfer while ensuring continuous and smooth height changes during lifting, translation, and descent through segmented interpolation, effectively avoiding equipment vibration or collision risks caused by sudden height changes. Moreover, through segmented planning and decoupling, the three-dimensional trajectory is split into planar and height dimensions for separate optimization, reducing the complexity of multi-dimensional coupled planning and improving the flexibility and accuracy of trajectory planning. Different spline curves of different orders are used in different stages, which not only meets the high precision requirements of welding within the workstation, but also takes into account the efficient obstacle avoidance requirements of transfer between workstations, thus achieving a balance between efficiency and precision. Furthermore, the trajectory can be dynamically adjusted by combining the theoretical coordinates of the weld point with the real-time offset vector, so as to adapt to the coordinate deviation in the actual welding scenario to the greatest extent, thereby effectively improving the robustness and real-time response capability of the system.

[0032] The coordinated scheduling unit includes: The timing management module is configured to generate and execute a time-triggered deterministic scheduling timing table, and through a hardware clock synchronization mechanism, ensure that the vision processing stage of the first station and the welding stage of the second station overlap in time during the cycle of alternating operations of the two stations, while maintaining phase locking in coordination with the mechanical vibration cycle to form a deterministic assembly line operation cycle. The status monitoring module is configured to periodically collect the trigger count of the image acquisition unit, the feature recognition confidence of the vision processing unit, the actual position and following error of each axis of the multi-axis control unit, and the voltage and current waveform data of the welding power supply. The module compares the above data with the preset process thresholds in real time, and generates a hierarchical alarm information log containing fault codes, timestamps and possible causes when the data exceeds the limit.

[0033] The technical effects of the above solution are as follows: The timing management module generates and executes a time-triggered deterministic scheduling timing table, and with the help of a hardware clock synchronization mechanism, it can ensure that the visual processing and welding stages of the two workstations are precisely overlapped in time, thus forming an efficient and deterministic production line operation rhythm. At the same time, its phase locking function, which is coordinated with the mechanical vibration cycle, effectively suppresses the interference of vibration on high-precision operations, thereby fundamentally improving the overall production efficiency and operational stability of the system. The status monitoring module periodically collects and compares the operating data and process thresholds of key links in the entire system (including image acquisition, visual recognition, motion control, and welding power supply), which can realize comprehensive real-time monitoring of the production process. In case of anomalies, it generates structured, traceable, hierarchical alarm logs, which not only enhances the system's fault early warning and diagnosis capabilities, but also provides accurate data support for welding process optimization and preventive maintenance, thereby ensuring the continuity of the production process and the consistency of product quality.

[0034] The collaborative scheduling unit also introduces a fault self-recovery mechanism, specifically including: Based on the received hierarchical alarm information, execute a tiered response strategy; For non-critical parameter drift alarms, the process parameters of subsequent cycles are fine-tuned according to the preset compensation rule library without interrupting the current work cycle. For critical motion and process fault alarms, immediately initiate a safety interruption sequence, freeze the motion of each axis, save the field status, and automatically execute the preset recovery process; After the recovery process is completed, based on the processing status and compensation data completed before the interruption, the execution timing and path of the remaining solder joints are recalculated and optimized to achieve seamless production recovery from the breakpoint. In subsequent scheduling, the identified fault modes are automatically avoided and the resource allocation strategy is dynamically adjusted.

[0035] In this embodiment, the specific implementation steps of automatically executing the preset recovery process are as follows: based on the type code of the critical fault alarm, the corresponding standardized recovery program script is matched and loaded from the preset fault recovery strategy library; the multi-axis control unit is controlled to raise the welding head to a safe height and move it to the equipment origin or safe standby position in sequence according to the safety sequence, while cutting off the output of the welding power supply; the image acquisition unit is instructed to re-photograph and reposition the faulty workstation to obtain the actual position status of the workpiece after the interruption, and compare and calibrate it with the field status data saved before the interruption; based on the calibrated position information, the completed weld point records, and the remaining weld point sequence, the movement path and execution sequence of the welding head are re-planned to ensure that the new path is seamlessly connected with the completed work and there is no risk of collision; after confirming that all axis positions are ready, welding parameters are reset, and the optical environment is re-inspected and qualified, a recovery ready signal is sent to the operation interface, and the system waits for authorization or automatic instructions to continue the welding operation from the interruption point.

[0036] The technical effects of the above solution are as follows: By executing a hierarchical response strategy based on hierarchical alarm information, the system can intelligently distinguish fault levels. Thus, for non-critical parameter drift alarms, online process fine-tuning can be performed based on the compensation rule base without interrupting the production cycle, maintaining production continuity. For critical faults, a safety interruption sequence is immediately initiated. By freezing movement, saving status, and executing preset recovery procedures, the safety of equipment and products is maximized. After the recovery process is completed, the timing and path of the remaining solder joints can be recalculated and optimized based on the processing status and compensation data before the interruption, achieving seamless production recovery from the breakpoint. Combined with the strategy of automatically avoiding identified fault modes and dynamically adjusting resource allocation in subsequent scheduling, the robustness, production efficiency, and intelligent operation and maintenance level of the system can be improved.

[0037] At the start of each scheduling cycle of the timing management module, the collaborative scheduling unit sends a synchronization command containing a global cycle number and a timestamp to the image acquisition unit, vision processing unit, multi-axis control unit, and welding power supply. It receives process data with timing marks from the above units and constructs a unified data frame corresponding to each workstation and each processing cycle in memory based on the global cycle number and timestamp. Before the status monitoring module performs real-time comparison, it performs logical consistency verification on the related data items in the unified data frame. If the verification fails, the data in that frame is discarded and a re-sampling mechanism for the specific data source is triggered. At the same time, interpolation prediction is performed based on the previous valid data frame to maintain the continuity of the current work cycle.

[0038] The technical effects of the above solution are as follows: By sending synchronization instructions with global cycle numbers and timestamps to all units of the entire system at the beginning of each scheduling cycle and receiving feedback data with time-series markers, a strictly time-aligned global data environment is constructed, thus laying a precise time reference for multi-unit collaboration. Furthermore, by constructing a unified data frame containing complete process information for each workstation and each processing cycle, and performing logical consistency verification before status monitoring, the reliability of the data source and the completeness of the system status description are ensured. When the verification fails, by discarding abnormal data frames, triggering a specific data source re-sampling mechanism, and combining it with the previous valid data frame for interpolation prediction, the interference of local data anomalies on the global system is effectively isolated. While ensuring data quality, the continuity of the current work cycle and the stability of the production process are maintained, thereby significantly improving the robustness and reliability of the entire dual-station straightening welding system.

[0039] Working Principle: Through a collaborative scheduling unit, a deterministic pipeline cycle of parallel alternation between two workstations is constructed. The image acquisition unit synchronously acquires images of the solder pads at both workstations. After compression by edge nodes, the images are processed in parallel by the vision processing unit, which can quickly calculate the position offset vector of each solder point. Based on this, the multi-axis control unit plans and drives the welding head to complete high-precision position compensation and welding according to the offset, combined with batch pre-compensation and compensation mechanisms for dynamic / static errors such as thermal deformation and vibration. Furthermore, when one workstation is performing vision processing, the other workstation is synchronously performing welding, realizing the overlap of processing and execution in time, thereby eliminating the waiting time of the traditional serial process. At the same time, the system ensures continuous operation through status monitoring and fault self-recovery mechanisms. This invention effectively overcomes the dynamic error of dual workstations and can deeply integrate vision correction into the work cycle, thereby significantly improving equipment capacity and cycle time while ensuring high precision and high reliability of welding.

[0040] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0041] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A correction system for a dual-station solder ball soldering machine based on machine vision and multi-axis collaboration, characterized in that, include: The image acquisition unit is configured to simultaneously acquire images of the workpiece pads on the first and second workstations. The vision processing unit is configured to receive and process the pad images of the first station and the second station in parallel, for identifying the actual position of the solder joints at each station and calculating the offset vector of the actual position of the solder joints at each station relative to the preset calibration position. The multi-axis control unit is configured to generate coordinated motion commands to drive the welding head for position compensation based on the received offset vector; The collaborative scheduling unit is configured to alternately trigger the image acquisition and processing flow of the first and second workstations according to a preset rhythm, and to schedule the multi-axis control unit to drive the welding head to perform position compensation and welding operations corresponding to the offset vectors on the first and second workstations in sequence, so as to realize the parallel and alternating operation of visual correction and welding operations of the two workstations.

2. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 1, characterized in that, The image acquisition unit includes: The synchronous control module receives unified acquisition instructions from the collaborative scheduling unit, generates two synchronous hardware trigger signals, and controls the industrial cameras set at the first and second workstations to acquire images respectively. The lighting adjustment module is configured to independently and dynamically adjust the intensity, spectrum, and incident angle of the light source set at the first and second workstations based on online monitoring of the reflectivity of the fixtures at different workstations and changes in ambient light.

3. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 1, characterized in that, The visual processing unit includes: The image preprocessing module is configured to use a parallel computing architecture to perform synchronous processing on the workpiece pad images of the two workstations, including grayscale processing, noise reduction processing and image enhancement processing. The image enhancement processing includes contrast stretching and histogram equalization. The feature recognition module is configured to locate and segment the solder joint region in the preprocessed workpiece solder pad image based on a trained solder joint recognition model. The solder joint recognition model is an image recognition model based on a deep convolutional neural network. The coordinate calculation module is configured to extract the geometric center coordinates of each solder joint area as the actual position, and convert the image pixel coordinates into actual physical coordinates in the machine motion coordinate system according to the preset camera calibration parameters, and then calculate the offset vector.

4. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 3, characterized in that, After calculating the offset vector, the vision processing unit performs statistical analysis on the offset vectors of the first and second workstations obtained in the continuous production cycle. If an overall offset pattern related to the batch clamping error of the workpiece is identified, global position compensation parameters are generated and sent to the multi-axis control unit for batch pre-compensation of the welding path of subsequent workpieces.

5. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 1, characterized in that, The multi-axis control unit includes: The motion planning module is configured to plan a smooth, singularity-free movement trajectory of the welding head in three-dimensional space based on the pre-stored weld point sequence coordinates in the welding path process library, the safe height of the welding head approaching and lifting, and the received offset vector, taking into account the acceleration and deceleration capabilities of each motion axis and the maximum jump constraint, and generating corresponding timing commands for the position, velocity, and acceleration of each axis. The position compensation module is configured to decompose the offset vector corresponding to each solder point into independent motion components of the X-axis, Y-axis, Z-axis and R-axis according to the machine kinematics model during the execution of the motion trajectory. The anti-collision monitoring module is configured to predict and prevent possible interference between the welding head and the fixture during the dual-station switching process by monitoring the current and position feedback of each axis motor.

6. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 5, characterized in that, Using a spline curve interpolation algorithm, a smooth, singularity-free trajectory for the welded joint in three-dimensional space is planned, including: After receiving the theoretical coordinates of the weld point sequence and the real-time offset vector, the complete welding path is automatically divided into two types of continuous sub-trajectory segments: the two types of continuous sub-trajectory segments include the welding segment within the workstation and the transfer segment between workstations. The welding segment within the workstation and the transfer segment between workstations included in the three-dimensional spatial motion trajectory of the welding head are decoupled into a planar trajectory and a height trajectory, respectively. The planar and height trajectories of the welding section within the workstation and the transfer section between workstations are smoothed using spline curve interpolation algorithms to obtain the smoothed planar and height trajectories of the welding section within the workstation and the transfer section between workstations. The planar and height trajectories of the welding section within the workstation and the transfer section between workstations after smoothing are integrated to form a smooth, singularity-free movement trajectory of the welding head in three-dimensional space.

7. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 6, characterized in that, The planar and height trajectories of the welding section within the workstation and the transfer section between workstations are smoothed using spline curve interpolation algorithms, respectively, to obtain the smoothed planar and height trajectories of the welding section within the workstation and the transfer section between workstations, including: Iterate through all the weld points contained in the welding section within the workstation in sequence; The trajectory segments corresponding to the planar trajectories between every two weld points in the welding section within the workstation are smoothed by using quintic spline interpolation. This process yields the smoothed trajectory segments corresponding to the planar trajectories between every two weld points in the welding section within the workstation. The height of the welding head remains constant throughout the trajectory segment corresponding to every two welding points within the welding section at the workstation. Iterate through all the solder joints contained in the transfer section between workstations in sequence; The trajectory segments corresponding to the planar trajectories between every two welding points in the planned obstacle avoidance path within the inter-station transfer section are smoothed sequentially using quintic spline curve interpolation. Cubic spline curve interpolation is used to smooth the trajectory segment corresponding to the height trajectory between every two welding points in the inter-station transfer section, thereby obtaining the smoothed trajectory segment corresponding to the height trajectory between every two welding points in the inter-station transfer section.

8. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 5, characterized in that, During execution, the motion planning module continuously collects actual motion trajectories, compensation effects, and welding quality data from historical welding cycles. It trains a strategy network using an online reinforcement learning algorithm to iteratively optimize the smoothness of the motion trajectory, velocity curve, and acceleration parameters. The position compensation module introduces an error compensation mechanism on the basis of performing feedforward compensation. It is used to identify and compensate online for low-frequency quasi-static errors caused by thermal deformation of the mechanical transmission chain and straightness error of the guide rail, as well as high-frequency dynamic errors caused by structural vibration due to alternating motion of the two workstations.

9. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 1, characterized in that, The coordinated scheduling unit includes: The timing management module is configured to generate and execute a time-triggered deterministic scheduling timing table, and through a hardware clock synchronization mechanism, ensure that the vision processing stage of the first station and the welding stage of the second station overlap in time during the cycle of alternating operations of the two stations, while maintaining a phase lock coordinated with the mechanical vibration cycle to form a deterministic assembly line operation cycle. The status monitoring module is configured to periodically collect the trigger count of the image acquisition unit, the feature recognition confidence of the vision processing unit, the actual position and following error of each axis of the multi-axis control unit, and the voltage and current waveform data of the welding power supply. The module compares the above data with the preset process thresholds in real time, and generates a hierarchical alarm information log containing fault codes, timestamps and possible causes when the data exceeds the limit.

10. The dual-station solder ball soldering machine correction system based on machine vision and multi-axis collaboration according to claim 9, characterized in that, The collaborative scheduling unit also introduces a fault self-recovery mechanism, specifically including: Based on the received hierarchical alarm information, execute a tiered response strategy; For non-critical parameter drift alarms, the process parameters of subsequent cycles are fine-tuned according to the preset compensation rule library without interrupting the current work cycle. For critical motion and process fault alarms, immediately initiate a safety interruption sequence, freeze the motion of each axis, save the field status, and automatically execute the preset recovery process; After the recovery process is completed, based on the processing status and compensation data completed before the interruption, the execution timing and path of the remaining solder joints are recalculated and optimized to achieve seamless production recovery from the breakpoint. In subsequent scheduling, the identified fault modes are automatically avoided and the resource allocation strategy is dynamically adjusted. At the start of each scheduling cycle of the timing management module, the collaborative scheduling unit sends a synchronization command containing a global cycle number and a timestamp to the image acquisition unit, vision processing unit, multi-axis control unit, and welding power supply. It receives process data with timing marks from the above units and constructs a unified data frame corresponding to each workstation and each processing cycle in memory based on the global cycle number and timestamp. Before the status monitoring module performs real-time comparison, it performs logical consistency verification on the related data items in the unified data frame. If the verification fails, the data in that frame is discarded and a re-sampling mechanism for a specific data source is triggered. At the same time, interpolation prediction is performed based on the previous valid data frame.

Images