Clutch collaborative robot automatic vision unloading method and system

Abstract

The application provides a clutch cooperative robot automatic vision feeding and discharging method and system, the method comprises the following steps: based on the parts to be fed and discharged corresponding to each working procedure in the whole process of clutch production, extracting the shape and position tolerance sensitive features of each part to be fed and discharged corresponding to the working procedure, synchronously binding each working procedure node to generate a corresponding dynamic feature template library; driving the collaborative robot to sequentially perform a review three-level positioning operation to determine the corresponding grabbing pose, synchronously combining the beat requirement of the target working procedure and the readiness state of the upstream and downstream stations on the production line to generate a corresponding initial feeding and discharging path; calling preset grabbing parameters and synchronously optimizing the initial feeding and discharging path based on the dynamic feature template library to generate a corresponding target feeding and discharging path; based on the preset grabbing parameters, controlling the collaborative robot to complete the corresponding grabbing task according to the target feeding and discharging path. The application can effectively reduce the cost of equipment customization and the debugging period.

Description

Technical Field

[0001] This invention relates to the field of [unclear - possibly related to a specific field], and particularly to an automatic vision-based loading and unloading method and system for a clutch collaborative robot. Background Technology

[0002] As the automotive manufacturing industry rapidly develops towards large-scale and high-precision production, the precision and efficiency requirements for the assembly and welding processes of clutches, as a core component of the automotive transmission system, continue to increase. Material loading and unloading, as a fundamental step in the entire clutch production process, directly determines the production cycle time and finished product quality. In traditional clutch production, the material loading and unloading process relies on manual operation, resulting in low efficiency, poor precision stability, and high labor intensity. This is no longer suitable for the high-volume, high-cycle production demands of modern automotive clutches, becoming a significant bottleneck restricting the industry's automation upgrade.

[0003] The rapid iteration of collaborative robots and machine vision technologies has provided mature technical support for the automated loading and unloading of automotive parts. Related solutions have been widely applied in parts processing scenarios, effectively improving the automation level of loading and unloading operations. However, existing technical solutions are mostly developed for single loading and unloading scenarios of general automotive parts, and are not well adapted to the specific production scenarios of clutches with multiple components, multiple processes, and high precision. An integrated solution for automated vision-based loading and unloading of clutches based on collaborative robots has not yet been formed, and the core pain points of clutch loading and unloading processes cannot be fully addressed.

[0004] Furthermore, the existing technology has the following core defects: First, the clutch components are diverse and complex, and the existing system lacks the flexibility to adapt, making it difficult to meet the operational needs of mass production of multiple components; second, it mostly adopts a single robot operation mode, lacking a multi-collaborative robot scheduling mechanism, which makes it impossible to achieve the rhythm matching and efficient parallel operation of multi-process loading and unloading; third, the unloading process does not integrate an online visual inspection module that is linked with the loading and unloading actions, making it impossible to identify and screen welding defects in real time, which can easily lead to the outflow of defective products; fourth, the integration of the vision module and the robot control system is low, and the functions of part recognition, defect detection and visual guidance are disconnected, making it impossible to achieve a deep integration of automated loading and unloading and intelligent quality inspection. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to provide an automatic vision loading and unloading method and system for clutch collaborative robots, so as to solve the problem that the existing technologies are mostly developed for single loading and unloading scenarios of general automotive parts, resulting in insufficient adaptability to specific production scenarios of clutches with multiple parts, multiple processes and high precision.

[0006] The first aspect of the present invention proposes:

[0007] An automatic vision-based loading and unloading method for a clutch collaborative robot, wherein the method includes:

[0008] Based on the components to be loaded and unloaded corresponding to each process in the clutch production process, the form and position tolerance sensitive features of each component to be loaded and unloaded are extracted and simultaneously bound to each process node to generate a corresponding dynamic feature template library.

[0009] When the loading and unloading command of the target process is triggered, the collaborative robot is driven to perform three-level positioning operations in sequence: global coarse positioning, local fine positioning, and tolerance feature verification, in order to determine the corresponding grasping pose. Simultaneously, combined with the cycle time requirements of the target process and the ready status of the upstream and downstream workstations on the production line, the corresponding initial loading and unloading path is generated.

[0010] Based on the structure and material characteristics of the component corresponding to the target process, preset capture parameters are retrieved, and the initial loading and unloading path is optimized simultaneously based on the dynamic feature template library to generate the corresponding target loading and unloading path.

[0011] Based on the preset grasping parameters, the collaborative robot is controlled to complete the corresponding grasping task according to the target loading and unloading path.

[0012] The beneficial effects of this invention are as follows: This technical solution effectively solves the core pain point of insufficient adaptability of existing general automotive parts loading and unloading technology to the multi-component, multi-process, and high-precision production scenarios of clutches. By constructing a dynamic template library of form and position tolerance sensitive features bound to process nodes, flexible adaptation to the entire clutch production process loading and unloading scenarios is achieved. The adoption of three-level positioning operation significantly improves the positioning accuracy of loading and unloading, avoiding assembly deviations and product defects caused by the form and position tolerances of high-precision components. At the same time, by combining production line cycle time, upstream and downstream workstation status, and component characteristics, the loading and unloading path and grasping parameters are optimized, which not only improves the efficiency of production line collaborative operation, but also avoids damage such as scratches and deformation during the component loading and unloading process, realizing automated, high-precision, and highly flexible operation of clutch production loading and unloading links.

[0013] Furthermore, the step of extracting the form and position tolerance sensitive features of each of the components to be loaded and unloaded corresponding to the process, and synchronously binding each process node to generate the corresponding dynamic feature template library includes:

[0014] Multi-dimensional tolerance decomposition is performed on the components to be loaded and unloaded in each process of the clutch, so that the dimensional tolerance, position tolerance, and attitude tolerance are mapped into visually recognizable feature vectors respectively. Simultaneously, based on the processing risk level of the process, tolerance-sensitive weights are assigned to different feature vectors to eliminate redundant features and form lightweight feature primitives.

[0015] The lightweight feature primitives are strongly associated with the tooling fixtures, gripping postures, and allowable deviation ranges of the corresponding processes to construct a ternary association structure of process-feature primitives-constraints. The feature drift thresholds under different batches and working conditions are recorded simultaneously to form an adaptive correction benchmark.

[0016] Based on the ternary correlation structure and the adaptive correction benchmark, a dynamic feature template library that can be automatically loaded as the process switches is dynamically generated.

[0017] Furthermore, the step of dynamically generating the dynamic feature template library that can be automatically loaded during process switching based on the ternary association structure and the adaptive correction benchmark includes:

[0018] Based on the ternary association structure, a process-feature template mapping index is constructed to assign a corresponding template identifier to each process node according to the mapping index. Simultaneously, the feature drift threshold in the adaptive correction benchmark is embedded into the template identifier to output the corresponding target feature data.

[0019] Based on the target feature data, the feature primitives in the ternary association structure are compared with the adaptive correction benchmark to dynamically generate a template fit score.

[0020] The optimal feature template is selected based on the template fit score, and the optimal feature template is dynamically updated simultaneously to generate the corresponding dynamic feature template library.

[0021] Furthermore, the step of generating the corresponding initial loading and unloading paths by combining the cycle time requirements of the target process and the readiness status of upstream and downstream workstations on the production line includes:

[0022] The cycle time requirement of the target process is broken down into rigid time window constraints for the entire process of grabbing, transferring, and unloading. The actual ready status, preset action sequence and dynamic space occupancy information of the upstream and downstream workstations on the production line are collected simultaneously and transformed into dynamic space access constraints for path execution.

[0023] Based on the dynamic spatial access constraints, combined with the grasping pose and the form and position tolerance sensitive features, a multi-dimensional coupled path feasible region is constructed. Simultaneously, based on the path feasible region, the entire loading and unloading process is decomposed into a grasping and positioning section, a smooth transfer section, and a precise unloading section.

[0024] Differentiated constraint priorities are matched to the grabbing and positioning section, the smooth transfer section, and the precise unloading section. Simultaneously, the initial path planning is completed based on the timing matching of each constraint priority to generate the initial loading and unloading path.

[0025] Furthermore, the step of performing initial path planning based on the priority of each constraint to generate the initial loading and unloading path includes:

[0026] Based on the respective constraint priorities, hierarchical and quantified constraint rules are constructed for the grasping and positioning segment, the smooth transfer segment, and the precise material release segment, respectively, to form a hierarchical constraint benchmark for path planning.

[0027] Based on the aforementioned hierarchical constraint benchmark, a time-coupled segmented progressive planning algorithm is adopted to sequentially generate the initial path for each segment, and simultaneously set smooth transition constraint nodes at the junctions of adjacent segments.

[0028] Each of the smooth transition constraint nodes is subjected to closed-loop verification to generate the corresponding initial loading and unloading path.

[0029] Furthermore, the step of optimizing the initial loading / unloading path based on the dynamic feature template library to generate the corresponding target loading / unloading path includes:

[0030] In the dynamic feature template library, the form and position tolerance sensitive features, tolerance allowable thresholds and feature verification requirements of the component corresponding to the target process are retrieved, and combined with the preset capture parameters, they are converted into forced feature anchor points on the path;

[0031] The spatial position, robot posture, and triggering sequence corresponding to each of the forced feature anchor points are detected to integrate the corresponding initial optimized path;

[0032] The initial optimized path is subjected to closed-loop checks for constraint compliance, cycle time adaptability, and spatial non-interference, in order to output the target loading and unloading path accordingly.

[0033] Furthermore, the step of performing closed-loop checks on the initial optimized path for constraint compliance, cycle time adaptability, and spatial non-interference to output the target loading / unloading path includes:

[0034] Based on the dynamic feature template library, a digital twin verification model that maps to the actual physical production line is constructed. Simultaneously, the initial optimized path and the forced feature anchor points are fully mapped into the interior of the digital twin verification model to build a virtual verification carrier.

[0035] Based on the virtual verification carrier, the initial optimized path is verified in a closed loop, and the deviation between the actual running data and the twin verification data is compared synchronously.

[0036] The initial optimized path is reviewed and verified based on the deviation to generate the target loading and unloading path accordingly.

[0037] The second aspect of the present invention proposes:

[0038] An automated vision-based loading and unloading system for a clutch collaborative robot, wherein the system includes:

[0039] The extraction module is used to extract the form and position tolerance sensitive features of each process corresponding to the parts to be loaded and unloaded in each process of the clutch production process, and to bind each process node in a synchronous manner to generate a corresponding dynamic feature template library.

[0040] The drive module is used to drive the collaborative robot to perform three levels of positioning operations in sequence: global coarse positioning, local fine positioning, and tolerance feature verification when the loading and unloading command of the target process is triggered, so as to determine the corresponding grasping posture and generate the corresponding initial loading and unloading path in combination with the cycle time requirements of the target process and the ready status of the upstream and downstream workstations on the production line.

[0041] The processing module is used to retrieve preset capture parameters based on the structure and material characteristics of the component corresponding to the target process, and simultaneously optimize the initial loading and unloading path based on the dynamic feature template library to generate the corresponding target loading and unloading path.

[0042] The control module is used to control the collaborative robot to complete the corresponding grasping task according to the target loading and unloading path based on the preset grasping parameters.

[0043] Furthermore, the extraction module is specifically used for:

[0044] Multi-dimensional tolerance decomposition is performed on the components to be loaded and unloaded in each process of the clutch, so that the dimensional tolerance, position tolerance, and attitude tolerance are mapped into visually recognizable feature vectors respectively. Simultaneously, based on the processing risk level of the process, tolerance-sensitive weights are assigned to different feature vectors to eliminate redundant features and form lightweight feature primitives.

[0045] The lightweight feature primitives are strongly associated with the tooling fixtures, gripping postures, and allowable deviation ranges of the corresponding processes to construct a ternary association structure of process-feature primitives-constraints. The feature drift thresholds under different batches and working conditions are recorded simultaneously to form an adaptive correction benchmark.

[0046] Based on the ternary correlation structure and the adaptive correction benchmark, a dynamic feature template library that can be automatically loaded as the process switches is dynamically generated.

[0047] Furthermore, the extraction module is specifically used for:

[0048] Based on the ternary association structure, a process-feature template mapping index is constructed to assign a corresponding template identifier to each process node according to the mapping index. Simultaneously, the feature drift threshold in the adaptive correction benchmark is embedded into the template identifier to output the corresponding target feature data.

[0049] Based on the target feature data, the feature primitives in the ternary association structure are compared with the adaptive correction benchmark to dynamically generate a template fit score.

[0050] The optimal feature template is selected based on the template fit score, and the optimal feature template is dynamically updated simultaneously to generate the corresponding dynamic feature template library.

[0051] Furthermore, the driving module is specifically used for:

[0052] The cycle time requirement of the target process is broken down into rigid time window constraints for the entire process of grabbing, transferring, and unloading. The actual ready status, preset action sequence and dynamic space occupancy information of the upstream and downstream workstations on the production line are collected simultaneously and transformed into dynamic space access constraints for path execution.

[0053] Based on the dynamic spatial access constraints, combined with the grasping pose and the form and position tolerance sensitive features, a multi-dimensional coupled path feasible region is constructed. Simultaneously, based on the path feasible region, the entire loading and unloading process is decomposed into a grasping and positioning section, a smooth transfer section, and a precise unloading section.

[0054] Differentiated constraint priorities are matched to the grabbing and positioning section, the smooth transfer section, and the precise unloading section. Simultaneously, the initial path planning is completed based on the timing matching of each constraint priority to generate the initial loading and unloading path.

[0055] Furthermore, the driving module is specifically used for:

[0056] Based on the respective constraint priorities, hierarchical and quantified constraint rules are constructed for the grasping and positioning segment, the smooth transfer segment, and the precise material release segment, respectively, to form a hierarchical constraint benchmark for path planning.

[0057] Based on the aforementioned hierarchical constraint benchmark, a time-coupled segmented progressive planning algorithm is adopted to sequentially generate the initial path for each segment, and simultaneously set smooth transition constraint nodes at the junctions of adjacent segments.

[0058] Each of the smooth transition constraint nodes is subjected to closed-loop verification to generate the corresponding initial loading and unloading path.

[0059] Furthermore, the processing module is specifically used for:

[0060] In the dynamic feature template library, the form and position tolerance sensitive features, tolerance allowable thresholds and feature verification requirements of the component corresponding to the target process are retrieved, and combined with the preset capture parameters, they are converted into forced feature anchor points on the path;

[0061] The spatial position, robot posture, and triggering sequence corresponding to each of the forced feature anchor points are detected to integrate the corresponding initial optimized path;

[0062] The initial optimized path is subjected to closed-loop checks for constraint compliance, cycle time adaptability, and spatial non-interference, in order to output the target loading and unloading path accordingly.

[0063] Furthermore, the processing module is specifically used for:

[0064] Based on the dynamic feature template library, a digital twin verification model that maps to the actual physical production line is constructed. Simultaneously, the initial optimized path and the forced feature anchor points are fully mapped into the interior of the digital twin verification model to build a virtual verification carrier.

[0065] Based on the virtual verification carrier, the initial optimized path is verified in a closed loop, and the deviation between the actual running data and the twin verification data is compared synchronously.

[0066] The initial optimized path is reviewed and verified based on the deviation to generate the target loading and unloading path accordingly.

[0067] The third aspect of the present invention proposes:

[0068] A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the automatic vision loading and unloading method for a clutch collaborative robot as described above.

[0069] The fourth aspect of the present invention proposes:

[0070] A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the automatic vision loading and unloading method for a clutch collaborative robot as described above.

[0071] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0072] Figure 1 A flowchart of the automatic vision loading and unloading method for a clutch cooperative robot provided in the first embodiment of the present invention;

[0073] Figure 2This is a structural block diagram of the automatic vision loading and unloading system for a clutch cooperative robot provided in the third embodiment of the present invention.

[0074] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0075] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0076] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0077] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0078] Please see Figure 1 The figure shows the automatic visual loading and unloading method for clutch collaborative robots provided in the first embodiment of the present invention. The automatic visual loading and unloading method for clutch collaborative robots provided in this embodiment can ultimately realize the automation, high precision and high flexibility of the clutch production loading and unloading process, effectively reduce the cost of equipment customization and modification and the debugging cycle, and fully meet the exclusive production needs of clutches.

[0079] Specifically, this embodiment provides:

[0080] An automatic vision-based loading and unloading method for a clutch collaborative robot, wherein the method includes:

[0081] Step S10: Based on the parts to be loaded and unloaded corresponding to each process in the clutch production process, extract the form and position tolerance sensitive features of each process corresponding to the parts to be loaded and unloaded, and bind each process node simultaneously to generate the corresponding dynamic feature template library.

[0082] It should be noted that, firstly, the entire clutch production process encompasses dozens of core processes, including stamping, turning, grinding, heat treatment, press fitting, assembly, and inspection. The structural characteristics, processing requirements, and geometric tolerance constraints of the components to be loaded and unloaded in different processes (such as clutch covers, pressure plates, diaphragm springs, driven plate assemblies, release bearings, etc.) are fundamentally different. Even for the same component, the tolerance-sensitive characteristics in different processes are completely different. For example, the flatness and parallelism of the pressure plate in the grinding process are core sensitive characteristics that directly affect the subsequent assembly quality, while the outer contour dimensions in the transfer process are the core control characteristics. Traditional visual loading and unloading uses fixed universal templates, which cannot distinguish the tolerance sensitivity priorities of different processes. This easily leads to problems such as missed detection of key tolerance characteristics, positioning deviations caused by redundant identification of non-core characteristics, and cycle delays. This step first precisely extracts the form and position tolerance sensitive features that play a decisive role in the processing quality and assembly accuracy of the parts to be loaded and unloaded in each process. At the same time, the features are strongly bound to the corresponding process nodes, laying the core data foundation for rapid template calling, accurate visual recognition, and positioning accuracy control during subsequent process switching. The dynamic feature template library realizes seamless adaptation during process changeover, eliminating the need for manual template recalibration and greatly improving the efficiency of production line changeover.

[0083] Step S20: When the loading and unloading command of the target process is triggered, the collaborative robot is driven to perform three-level positioning operations in sequence: global coarse positioning, local fine positioning, and tolerance feature verification, so as to determine the corresponding grasping pose. Simultaneously, the initial loading and unloading path is generated by combining the cycle time requirements of the target process and the ready status of the upstream and downstream workstations on the production line.

[0084] It's important to note that the core challenge in clutch loading and unloading lies in the fact that the accuracy of the gripping posture directly impacts the form and position tolerance pass rate of subsequent processing. Traditional single-level vision positioning either offers fast global positioning but insufficient accuracy, or high local positioning accuracy but limited coverage, failing to balance production line cycle time and gripping accuracy. This step employs a three-level progressive positioning logic: global coarse positioning quickly locks the overall position and approximate posture of the parts to be loaded or unloaded using wide-angle vision, enabling rapid robot positioning and significantly reducing time lost on large-scale movements; local fine positioning focuses on the core feature areas of the parts using high-precision close-range vision, accurately calculating the spatial coordinates and posture angles for gripping, ensuring basic gripping accuracy; tolerance feature verification performs a final check on the form and position tolerance sensitive features of this process, ensuring that the gripping posture will not damage critical tolerance surfaces of the parts or affect the clamping and positioning accuracy of subsequent processes. This three-level progressive positioning perfectly balances production line cycle time requirements with gripping and positioning accuracy. Meanwhile, the clutch production line is a multi-station continuous flow production line. The cycle time and path of loading and unloading at a single station must be deeply coordinated with the upstream and downstream stations. Otherwise, problems such as station waiting, material accumulation, and even interference between upstream and downstream actions will occur. Therefore, this step simultaneously collects the cycle time requirements of the target process (i.e., the longest execution time allowed for single-step loading and unloading) and the readiness status of upstream and downstream stations (such as whether the upstream process has completed processing and unloading, and whether the downstream process has completed tooling clearing and is ready to receive materials) while determining the gripping posture. Based on this, the initial loading and unloading path is generated to ensure that the path planning is in line with the actual flow needs of the production line from the source, and to avoid the path from being out of sync with the production line cycle time and station status.

[0085] Step S30: Based on the structure and material characteristics of the component corresponding to the target process, retrieve the preset capture parameters, and simultaneously optimize the initial loading and unloading path based on the dynamic feature template library to generate the corresponding target loading and unloading path;

[0086] It should be noted that the structures (such as thin-walled diaphragm springs, heavy pressure plates, and splined driven plates) and materials (such as stamped steel plates, cast iron, and friction materials) of different clutch components vary greatly. Consequently, the corresponding gripping parameters (such as gripper opening and closing stroke, gripping force, adsorption pressure, and approach speed) are completely different. For example, the diaphragm spring is a thin-walled elastic component. Excessive gripping force can lead to plastic deformation, while insufficient force may result in the risk of it falling off. Therefore, it is necessary to pre-set gripping parameters that match the structure and material of the components to ensure the safety and stability of the gripping. Meanwhile, the initial loading and unloading paths only met the basic requirements of cycle time and workstation coordination, without fully considering the constraints of the form and position tolerance sensitive features of the target process. For example, for the pressure plate in the grinding process, the path needs to ensure that it will not touch the machined high-precision plane when gripping, and it needs to accurately fit the positioning reference of the tooling when unloading. Therefore, this step optimizes the initial path based on the tolerance sensitive features, allowable deviation range, and feature verification requirements of the process in the dynamic feature template library. For example, the moving speed is reduced when approaching the high-precision tolerance surface, the attitude micro-adjustment point is added in the unloading section, and the space area that may touch the tolerance surface is avoided in the transfer section. The final target loading and unloading path not only meets the production line cycle time requirements, but also fully adapts to the tolerance control requirements of the target process, while taking into account the safety and stability of gripping.

[0087] Step S40: Based on the preset grasping parameters, control the collaborative robot to complete the corresponding grasping task according to the target loading and unloading path.

[0088] It should be noted that this step is the execution loop of the entire method. Based on the previously determined precise grasping pose, the adapted grasping parameters, and the optimized target path, the collaborative robot is controlled to complete the entire process of picking up, transferring, and placing materials. At the same time, during the execution process, the actual execution data can be synchronously fed back to the dynamic feature template library, providing data support for the dynamic updating of templates and the continuous optimization of subsequent paths, forming a closed-loop control of the entire process.

[0089] Second Embodiment

[0090] Furthermore, the step of extracting the form and position tolerance sensitive features of each of the components to be loaded and unloaded corresponding to the process, and synchronously binding each process node to generate the corresponding dynamic feature template library includes:

[0091] Multi-dimensional tolerance decomposition is performed on the components to be loaded and unloaded in each process of the clutch, so that the dimensional tolerance, position tolerance, and attitude tolerance are mapped into visually recognizable feature vectors respectively. Simultaneously, based on the processing risk level of the process, tolerance-sensitive weights are assigned to different feature vectors to eliminate redundant features and form lightweight feature primitives.

[0092] The lightweight feature primitives are strongly associated with the tooling fixtures, gripping postures, and allowable deviation ranges of the corresponding processes to construct a ternary association structure of process-feature primitives-constraints. The feature drift thresholds under different batches and working conditions are recorded simultaneously to form an adaptive correction benchmark.

[0093] Based on the ternary correlation structure and the adaptive correction benchmark, a dynamic feature template library that can be automatically loaded as the process switches is dynamically generated.

[0094] It should be noted that the form and position tolerances of clutch components cover three main categories: dimensional tolerances (such as inner and outer diameters and thickness), positional tolerances (such as coaxiality and positional tolerance), and orientation tolerances (such as parallelism, flatness, and perpendicularity). Different tolerances have different requirements for visual recognition. Traditional templates usually include all features indiscriminately in the recognition range, resulting in redundant feature dimensions, slow recognition speed, and the submergence of core tolerance features. This step first decomposes the tolerances of each component into multiple dimensions, transforming each tolerance requirement into a feature vector that can be recognized and quantified by the vision system. For example, the flatness tolerance of the pressure plate is transformed into a point cloud feature vector of the surface contour, and the coaxiality tolerance is transformed into a feature vector of the center distance between the inner and outer circles. At the same time, based on the processing risk level of the process, specifically the severity of the quality consequences caused by exceeding a certain tolerance, different tolerance sensitivity weights are assigned to different feature vectors. For example, exceeding the flatness tolerance in the grinding process will directly lead to the scrapping of the component, so it is assigned the highest weight, while the chamfer size of the component has no impact on the processing quality, so it is assigned the lowest weight or even directly eliminated. The resulting lightweight feature primitives retain the core sensitive features that play a decisive role in the process quality, while eliminating redundant non-core features, which greatly improves the speed and accuracy of subsequent visual recognition and avoids the interference of invalid features on positioning accuracy.

[0095] Visual recognition feature primitives do not exist in isolation; they must be deeply bound to the actual execution conditions of the process to ensure the template's usability. Specifically, for the same feature primitive, the optimal gripping posture and allowable recognition deviation range will be completely different under different tooling fixture clamping methods. Traditional templates only focus on the feature itself and are not bound to the actual constraints of the process, resulting in the need for manual adjustment of the gripping posture and deviation threshold after template usage, leading to extremely low template changeover efficiency. This step strongly associates lightweight feature primitives with the corresponding process's tooling fixture model, clamping and positioning method, optimal gripping posture, and maximum allowable deviation range, constructing a ternary association structure of "process - feature primitive - constraint conditions." This ensures that each feature template is fully adapted to the entire process execution requirements, and all constraints can be obtained synchronously by calling the template, eliminating the need for secondary manual adjustments. Meanwhile, the production of clutch components may involve batch-to-batch differences in raw materials and feature drift caused by wear of processing equipment. For example, pressure plates from the same batch may experience systematic slight drifts in their outer diameter due to wear of machine tool cutting tools. Traditional fixed templates cannot cope with this drift, resulting in a continuous decline in recognition success rate. Therefore, this step simultaneously records the feature drift thresholds under different batches and different operating conditions (such as different ambient temperatures and different equipment operating times) to form an adaptive correction benchmark, providing core data support for the dynamic updating and automatic adaptation of the template.

[0096] This step is the final closed loop of template library generation. Based on the previously constructed ternary association structure, a unique feature template is generated for each process. At the same time, based on the adaptive correction benchmark, automatic correction rules are set for the templates, so that the template library can automatically load the corresponding templates as the process changes, and can automatically correct the feature thresholds as the batch and working conditions change. This is completely different from the traditional fixed template library, realizing full dynamic adaptation of templates, and providing a stable and accurate visual recognition foundation for subsequent three-level positioning and path optimization.

[0097] Furthermore, the step of dynamically generating the dynamic feature template library that can be automatically loaded during process switching based on the ternary association structure and the adaptive correction benchmark includes:

[0098] Based on the ternary association structure, a process-feature template mapping index is constructed to assign a corresponding template identifier to each process node according to the mapping index. Simultaneously, the feature drift threshold in the adaptive correction benchmark is embedded into the template identifier to output the corresponding target feature data.

[0099] Based on the target feature data, the feature primitives in the ternary association structure are compared with the adaptive correction benchmark to dynamically generate a template fit score.

[0100] The optimal feature template is selected based on the template fit score, and the optimal feature template is dynamically updated simultaneously to generate the corresponding dynamic feature template library.

[0101] It's important to note that the clutch production line operates at high speeds. After a loading / unloading command is triggered, the corresponding template needs to be retrieved within milliseconds. Traditional template libraries without indexes often suffer from slow template retrieval and matching errors when dealing with a large number of processes. Therefore, this step utilizes a ternary association structure to construct a one-to-one mapping index between "process" and "feature template." Each process node is assigned a unique template identifier, allowing the production line to directly locate the corresponding feature template when a loading / unloading command for a particular process is triggered, significantly improving the speed and accuracy of template retrieval. Simultaneously, the feature drift threshold from the adaptive correction benchmark is embedded into the template identifier. This allows the feature correction parameters of the template to be retrieved simultaneously with the template identifier, eliminating the need for a secondary retrieval of the correction benchmark and further improving template loading efficiency. The output target feature data includes both core feature primitives and correction parameters adapted to the current operating conditions, providing accurate input for subsequent visual recognition.

[0102] During production line operation, the characteristics of components change with batches and operating conditions. It is necessary to determine the adaptability of the current template to the current component to avoid recognition errors caused by template mismatch. This step compares the actual features of the current component with the standard feature primitives in the ternary association structure based on the retrieved target feature data. At the same time, it combines the feature drift threshold in the adaptive correction benchmark to calculate the adaptability score of the current template to the current component. The higher the score, the higher the feature matching degree between the template and the current component, and the more guaranteed the recognition accuracy. When the adaptability score is lower than the preset threshold, the template is updated and optimized, thus avoiding recognition failures caused by template aging or mismatch from a mechanism perspective.

[0103] For the same process, there may be multiple alternative templates suitable for different batches and working conditions. This step selects the optimal feature template with the highest suitability for the current component based on the suitability score, ensuring the accuracy and success rate of visual recognition. At the same time, the actual operating data of the optimal feature template (such as actual recognition deviation, capture success rate, and working condition parameters) is fed back to the template library. The feature primitives, correction benchmarks, and constraints of the template are dynamically updated, so that the template library can be continuously optimized as the production line runs, always maintaining the highest suitability. The resulting dynamic feature template library has the core advantages of "fast call, accurate matching, automatic optimization, and dynamic updating", and is fully adaptable to the multi-process and variable working condition production scenarios of clutches.

[0104] Furthermore, the step of generating the corresponding initial loading and unloading paths by combining the cycle time requirements of the target process and the readiness status of upstream and downstream workstations on the production line includes:

[0105] The cycle time requirement of the target process is broken down into rigid time window constraints for the entire process of grabbing, transferring, and unloading. The actual ready status, preset action sequence and dynamic space occupancy information of the upstream and downstream workstations on the production line are collected simultaneously and transformed into dynamic space access constraints for path execution.

[0106] Based on the dynamic spatial access constraints, combined with the grasping pose and the form and position tolerance sensitive features, a multi-dimensional coupled path feasible region is constructed. Simultaneously, based on the path feasible region, the entire loading and unloading process is decomposed into a grasping and positioning section, a smooth transfer section, and a precise unloading section.

[0107] Differentiated constraint priorities are matched to the grabbing and positioning section, the smooth transfer section, and the precise unloading section. Simultaneously, the initial path planning is completed based on the timing matching of each constraint priority to generate the initial loading and unloading path.

[0108] It's important to note that the cycle time requirement for the clutch production line is rigid. Timeouts at a single workstation will directly disrupt the cycle time of the entire line. Traditional path planning typically only sets a total time constraint, failing to break it down into individual actions throughout the entire process. This makes it highly susceptible to problems where timeouts in a single step cause overall cycle time delays. This step first breaks down the total cycle time requirement of the target process into rigid time window constraints for the entire process: gripping and positioning, gripping action, smooth transfer, unloading action, and robot reset. A clear maximum allowable execution time is set for each action, ensuring that the path planning aligns with the cycle time requirement from the outset. Meanwhile, the upstream and downstream workstations of the clutch production line operate dynamically. For example, the pushing mechanism of the upstream process and the tooling chuck of the downstream process will occupy some space within a fixed sequence. If the loading and unloading paths do not take into account these dynamic space occupancy, there will be serious risks of mechanism interference and collision. Therefore, this step simultaneously collects the actual ready status, preset action sequence, and dynamic space occupancy information of the upstream and downstream workstations, and transforms them into dynamic space access constraints for path execution. Specifically, it clarifies which space areas are prohibited from entering during which time periods and which areas are safe to pass through, ensuring that the path planning is fully adapted to the dynamic operating status of the production line and avoiding interference risks from the source.

[0109] The feasible path domain is the spatial range within which a robot can safely perform actions. Traditional feasible path domains typically only consider single-dimensional constraints for spatial obstacle avoidance, without incorporating the posture requirements of the grasping position or the protection requirements for form and position tolerance-sensitive features. This results in planned paths that, while interference-free, may touch tolerance-sensitive surfaces of components or fail to meet the posture requirements for grasping and unloading. This step constructs a multi-dimensional coupled feasible path domain based on dynamic spatial access constraints, coordinate and posture requirements of the grasping position, and protection requirements for form and position tolerance-sensitive features (such as prohibiting entry into safe areas around tolerance surfaces). This ensures that the planned path remains entirely within a safe, compliant, and executable range. Meanwhile, the different stages of the entire loading and unloading process have completely different requirements for the robot's speed, accuracy, and posture. The gripping and positioning stage requires rapid and large-scale movement, which has high speed requirements and relatively low accuracy requirements. The smooth transfer stage needs to ensure the stability of the components and avoid material shaking or falling, which has high requirements for operational stability. The precise placement stage needs to conform to the tooling positioning reference, which has the highest requirements for accuracy and posture and low speed requirements. Therefore, this step breaks down the entire loading and unloading process into three distinct segments, laying the foundation for subsequent differentiated constraint settings and path planning, and solving the pain point that traditional single path planning cannot take into account speed, stability, and accuracy.

[0110] The three segments have different core requirements, and their corresponding constraint priorities are also completely different. The core requirement of the grabbing and positioning segment is rapid positioning, so the constraint priority is "cycle time constraint > spatial obstacle avoidance constraint > attitude accuracy constraint". The core requirement of the smooth transfer segment is smooth operation without interference, so the constraint priority is "spatial obstacle avoidance constraint > operational smoothness constraint > cycle time constraint". The core requirement of the precise material release segment is precise material release without damaging tolerance surfaces, so the constraint priority is "form and position tolerance constraint > attitude accuracy constraint > spatial obstacle avoidance constraint > cycle time constraint". This step matches the differentiated constraint priorities of the three segments, allowing path planning to focus on the core requirements of different segments and avoiding the problem of neglecting some aspects due to traditional uniform constraints. At the same time, based on the constraint priorities, the initial path planning is completed to match the timing requirements and spatial constraints. The generated initial loading and unloading paths not only meet the cycle time requirements of the production line and the coordination requirements of upstream and downstream workstations, but also take into account the core requirements of different segments, providing a reliable foundation for subsequent path optimization.

[0111] Furthermore, the step of performing initial path planning based on the priority of each constraint to generate the initial loading and unloading path includes:

[0112] Based on the respective constraint priorities, hierarchical and quantified constraint rules are constructed for the grasping and positioning segment, the smooth transfer segment, and the precise material release segment, respectively, to form a hierarchical constraint benchmark for path planning.

[0113] Based on the aforementioned hierarchical constraint benchmark, a time-coupled segmented progressive planning algorithm is adopted to sequentially generate the initial path for each segment, and simultaneously set smooth transition constraint nodes at the junctions of adjacent segments.

[0114] Each of the smooth transition constraint nodes is subjected to closed-loop verification to generate the corresponding initial loading and unloading path.

[0115] It's important to note that differentiated constraint priorities need to be transformed into quantifiable and executable constraint rules in order to be recognized and executed by the path planning algorithm. Traditional path planning constraints are usually qualitative and cannot achieve hierarchical control. This step transforms each constraint into a hierarchically quantifiable rule based on the constraint priority of each segment. For example, for the grasping and positioning segment, the highest priority cycle time constraint is quantified as "maximum movement time not exceeding 0.8s, maximum running speed not exceeding 2m / s"; the second priority spatial obstacle avoidance constraint is quantified as "minimum safe distance from obstacles not less than 50mm"; and the lowest priority attitude accuracy constraint is quantified as "positioning attitude deviation not exceeding ±5°". For the precise material feeding segment, the highest priority form and position tolerance constraint is quantified as "the parallelism deviation between the feeding posture and the tooling datum does not exceed 0.02mm"; the second priority posture accuracy constraint is quantified as "the feeding position deviation does not exceed ±0.05mm", and so on. A hierarchical quantified constraint rule is constructed for each segment, forming a hierarchical constraint datum for path planning. This allows the path planning algorithm to accurately execute the constraint requirements of each segment, ensuring that the path fully conforms to the core requirements of each segment.

[0116] Segmented path planning requires ensuring the temporal continuity and motion consistency of successive segments. Traditional segmented planning typically involves planning each segment independently and then forcibly splicing them together, which easily leads to problems such as sudden speed changes, posture jumps, and timing misalignments at the transition points. This results in significant impacts during robot operation, material swaying, and even timeouts. This step employs a temporally coupled segmented progressive planning algorithm. First, the path planning for the previous segment is completed. Then, using the endpoint coordinates, endpoint posture, endpoint speed, and endpoint timing of the previous segment as initial conditions, the path planning for the next segment is performed, ensuring complete coupling of the temporal and motion sequences of successive segments. Simultaneously, smooth transition constraint nodes are set at the transition points between adjacent segments to smoothly constrain the speed, acceleration, and jerk at the transition points, avoiding sudden changes in motion parameters, ensuring the stability of robot operation, and preventing material drop and component damage caused by impacts.

[0117] Setting smooth transition constraint nodes requires ensuring that the motion parameters at the connection point meet the smoothness requirements without violating the constraint rules of each segment. Therefore, this step performs closed-loop verification on all smooth transition constraint nodes. The verification includes whether the velocity and acceleration at the connection point are within the allowable range of the hierarchical constraint benchmark, whether the timing meets the requirements of the rigid time window, and whether the spatial position is within the feasible domain of the path. Only when all constraint nodes pass the verification can it be confirmed that the connection of the segmented path is compliant and executable. The final generated initial loading and unloading path is the basic path with full-process timing coupling, smooth motion, compliant constraints, and direct execution.

[0118] Furthermore, the step of optimizing the initial loading / unloading path based on the dynamic feature template library to generate the corresponding target loading / unloading path includes:

[0119] In the dynamic feature template library, the form and position tolerance sensitive features, tolerance allowable thresholds and feature verification requirements of the component corresponding to the target process are retrieved, and combined with the preset capture parameters, they are converted into forced feature anchor points on the path;

[0120] The spatial position, robot posture, and triggering sequence corresponding to each of the forced feature anchor points are detected to integrate the corresponding initial optimized path;

[0121] The initial optimized path is subjected to closed-loop checks for constraint compliance, cycle time adaptability, and spatial non-interference, in order to output the target loading and unloading path accordingly.

[0122] It should be noted that the initial loading and unloading path only meets the basic requirements of cycle time, space, and stability, and does not incorporate the control requirements of form and position tolerance sensitive features into the key nodes of the path. For example, for the pressure plate after high-precision grinding, the flatness of the pressure plate needs to be checked before unloading, and corresponding check stopping nodes need to be set on the path. At the same time, it is necessary to ensure that the high-precision surface that has been processed will not be touched during the gripping process, and corresponding obstacle avoidance anchor points need to be set on the path. This step retrieves the form and position tolerance sensitive features, tolerance thresholds, and feature verification requirements corresponding to the target process from the dynamic feature template library. Combined with preset gripping parameters (such as the opening and closing sequence of the gripper and the nodes of gripping force change), these constraints are transformed into mandatory feature anchor points on the path. Specifically, these anchor points are fixed nodes that must be passed through on the path and must meet the corresponding constraints, such as tolerance feature verification anchor points, high-precision surface obstacle avoidance anchor points, gripping force adjustment anchor points, and material release posture fine-tuning anchor points. The mandatory feature anchor points extend the tolerance control requirements from the visual recognition stage to the entire process of robot movement, realizing full-link coverage of tolerance control.

[0123] After the mandatory feature anchors are set, they need to be integrated into the initial path to form a complete optimized path. This step first accurately detects the spatial coordinates, robot posture requirements, and trigger timing nodes corresponding to each mandatory feature anchor. Then, using these anchors as key nodes, the initial path is refitted and adjusted to ensure that the optimized path passes through all mandatory feature anchors in sequence, and that the robot posture, trigger timing, and motion parameters at each anchor fully meet the constraint requirements. This integrates to form the initial optimized path, which has fully integrated the core requirements of tolerance control into the robot's motion trajectory.

[0124] After incorporating mandatory feature anchors into the initial optimized path, it needs to be re-verified to ensure it meets all constraint requirements, preventing the path from exceeding existing constraints due to the addition of anchors. This step involves three closed-loop checks on the initial optimized path: first, constraint compliance check, verifying whether all nodes on the path meet the requirements of geometric tolerance constraints, capture parameter constraints, and hierarchical constraint rules; second, cycle time adaptability check, verifying whether the execution time of the entire process meets the rigid time window requirements and whether it matches the timing of upstream and downstream workstations; and third, spatial non-interference check, verifying whether the entire path is within the feasible domain and whether there is any risk of interference with production line equipment or upstream and downstream mechanisms. Only when all three checks pass can the initial optimized path be determined as the final target loading and unloading path, ensuring that the optimized path meets both the high-precision requirements of tolerance control and the cycle time requirements and operational safety of the production line.

[0125] Furthermore, the step of performing closed-loop checks on the initial optimized path for constraint compliance, cycle time adaptability, and spatial non-interference to output the target loading / unloading path includes:

[0126] Based on the dynamic feature template library, a digital twin verification model that maps to the actual physical production line is constructed. Simultaneously, the initial optimized path and the forced feature anchor points are fully mapped into the interior of the digital twin verification model to build a virtual verification carrier.

[0127] Based on the virtual verification carrier, the initial optimized path is verified in a closed loop, and the deviation between the actual running data and the twin verification data is compared synchronously.

[0128] The initial optimized path is reviewed and verified based on the deviation to generate the target loading and unloading path accordingly.

[0129] It should be noted that the equipment layout, tooling fixtures, robot models, and motion sequences of upstream and downstream mechanisms in the clutch production line are extremely complex. Real-world debugging is not only time-consuming and labor-intensive but also carries high risks of collisions, equipment damage, and material scrap, especially for paths with high precision tolerances, where real-world debugging cannot accurately predict all hidden interferences and risks. This step first uses process information, tooling fixture information, and component feature information from a dynamic feature template library, combined with the physical production line's equipment layout, robot parameters, and motion sequences of upstream and downstream mechanisms, to construct a digital twin verification model that is precisely mapped 1:1 to the physical production line. This model can completely reproduce the physical production line's operating state, spatial layout, and motion logic. Then, the initial optimized path and all forced feature anchor points are fully mapped into the twin model, building a virtual verification carrier that is completely consistent with the physical scene, providing a safe, efficient, and accurate virtual environment for subsequent closed-loop verification.

[0130] In the virtual calibration platform, the entire process of a robot executing loading and unloading actions according to the initial optimized path can be fully simulated. The calibration content includes not only conventional spatial interference and cycle time adaptability, but also a full range of aspects such as robot motion parameters, joint torque, material stability, tolerance feature protection effect, and execution accuracy of forced feature anchor points. It can accurately identify problems such as implicit interference, motion impact, cycle time deviation, and posture deviation that cannot be predicted during actual machine debugging. At the same time, during the simulation, the running data of the twin calibration (such as the robot's actual running time, posture accuracy of each node, and distance to obstacles) is collected synchronously and compared with the theoretical data of path planning. The deviation between the two is calculated, providing accurate data support for subsequent path optimization.

[0131] Based on the deviation data obtained from the twin verification, the initial optimized path is specifically corrected and optimized. For example, for deviations that exceed the runtime limit, the running speed of the transfer section is optimized; for deviations that exceed the attitude accuracy limit, the attitude fine-tuning points of the unloading section are adjusted; for deviations with implicit interference, the spatial trajectory of the path is adjusted. After the correction is completed, it is re-verified in the twin verification model until the path fully meets all constraints and all deviations are within the allowable range. The final target loading and unloading path is the optimal path that has undergone full-scene virtual verification, is interference-free and risk-free, and perfectly matches the actual operating state of the production line. It can be directly issued to the robot for execution, which greatly reduces the risk and cost of actual machine debugging and improves the first-time success rate and operational stability of loading and unloading.

[0132] Please see Figure 2 The third embodiment of the present invention provides:

[0133] An automated vision-based loading and unloading system for a clutch collaborative robot, wherein the system includes:

[0134] The extraction module is used to extract the form and position tolerance sensitive features of each process corresponding to the parts to be loaded and unloaded in each process of the clutch production process, and to bind each process node in a synchronous manner to generate a corresponding dynamic feature template library.

[0135] The drive module is used to drive the collaborative robot to perform three levels of positioning operations in sequence: global coarse positioning, local fine positioning, and tolerance feature verification when the loading and unloading command of the target process is triggered, so as to determine the corresponding grasping posture and generate the corresponding initial loading and unloading path in combination with the cycle time requirements of the target process and the ready status of the upstream and downstream workstations on the production line.

[0136] The processing module is used to retrieve preset capture parameters based on the structure and material characteristics of the component corresponding to the target process, and simultaneously optimize the initial loading and unloading path based on the dynamic feature template library to generate the corresponding target loading and unloading path.

[0137] The control module is used to control the collaborative robot to complete the corresponding grasping task according to the target loading and unloading path based on the preset grasping parameters.

[0138] Furthermore, the extraction module is specifically used for:

[0139] Multi-dimensional tolerance decomposition is performed on the components to be loaded and unloaded in each process of the clutch, so that the dimensional tolerance, position tolerance, and attitude tolerance are mapped into visually recognizable feature vectors respectively. Simultaneously, based on the processing risk level of the process, tolerance-sensitive weights are assigned to different feature vectors to eliminate redundant features and form lightweight feature primitives.

[0140] The lightweight feature primitives are strongly associated with the tooling fixtures, gripping postures, and allowable deviation ranges of the corresponding processes to construct a ternary association structure of process-feature primitives-constraints. The feature drift thresholds under different batches and working conditions are recorded simultaneously to form an adaptive correction benchmark.

[0141] Based on the ternary correlation structure and the adaptive correction benchmark, a dynamic feature template library that can be automatically loaded as the process switches is dynamically generated.

[0142] Furthermore, the extraction module is specifically used for:

[0143] Based on the ternary association structure, a process-feature template mapping index is constructed to assign a corresponding template identifier to each process node according to the mapping index. Simultaneously, the feature drift threshold in the adaptive correction benchmark is embedded into the template identifier to output the corresponding target feature data.

[0144] Based on the target feature data, the feature primitives in the ternary association structure are compared with the adaptive correction benchmark to dynamically generate a template fit score.

[0145] The optimal feature template is selected based on the template fit score, and the optimal feature template is dynamically updated simultaneously to generate the corresponding dynamic feature template library.

[0146] Furthermore, the driving module is specifically used for:

[0147] The cycle time requirement of the target process is broken down into rigid time window constraints for the entire process of grabbing, transferring, and unloading. The actual ready status, preset action sequence and dynamic space occupancy information of the upstream and downstream workstations on the production line are collected simultaneously and transformed into dynamic space access constraints for path execution.

[0148] Based on the dynamic spatial access constraints, combined with the grasping pose and the form and position tolerance sensitive features, a multi-dimensional coupled path feasible region is constructed. Simultaneously, based on the path feasible region, the entire loading and unloading process is decomposed into a grasping and positioning section, a smooth transfer section, and a precise unloading section.

[0149] Differentiated constraint priorities are matched to the grabbing and positioning section, the smooth transfer section, and the precise unloading section. Simultaneously, the initial path planning is completed based on the timing matching of each constraint priority to generate the initial loading and unloading path.

[0150] Furthermore, the driving module is specifically used for:

[0151] Based on the respective constraint priorities, hierarchical and quantified constraint rules are constructed for the grasping and positioning segment, the smooth transfer segment, and the precise material release segment, respectively, to form a hierarchical constraint benchmark for path planning.

[0152] Based on the aforementioned hierarchical constraint benchmark, a time-coupled segmented progressive planning algorithm is adopted to sequentially generate the initial path for each segment, and simultaneously set smooth transition constraint nodes at the junctions of adjacent segments.

[0153] Each of the smooth transition constraint nodes is subjected to closed-loop verification to generate the corresponding initial loading and unloading path.

[0154] Furthermore, the processing module is specifically used for:

[0155] In the dynamic feature template library, the form and position tolerance sensitive features, tolerance allowable thresholds and feature verification requirements of the component corresponding to the target process are retrieved, and combined with the preset capture parameters, they are converted into forced feature anchor points on the path;

[0156] The spatial position, robot posture, and triggering sequence corresponding to each of the forced feature anchor points are detected to integrate the corresponding initial optimized path;

[0157] The initial optimized path is subjected to closed-loop checks for constraint compliance, cycle time adaptability, and spatial non-interference, in order to output the target loading and unloading path accordingly.

[0158] Furthermore, the processing module is specifically used for:

[0159] Based on the dynamic feature template library, a digital twin verification model that maps to the actual physical production line is constructed. Simultaneously, the initial optimized path and the forced feature anchor points are fully mapped into the interior of the digital twin verification model to build a virtual verification carrier.

[0160] Based on the virtual verification carrier, the initial optimized path is verified in a closed loop, and the deviation between the actual running data and the twin verification data is compared synchronously.

[0161] The initial optimized path is reviewed and verified based on the deviation to generate the target loading and unloading path accordingly.

[0162] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the automatic visual loading and unloading method for a clutch collaborative robot as described above.

[0163] The fifth embodiment of the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the automatic visual loading and unloading method for a clutch collaborative robot as described above.

[0164] In summary, the automatic vision loading and unloading method and system for clutch collaborative robots provided in the above embodiments of the present invention can ultimately realize the automation, high precision, and high flexibility of clutch production loading and unloading processes, effectively reducing equipment customization and modification costs and debugging cycles, and fully meeting the specific production needs of clutches.

[0165] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0166] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0167] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0168] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0169] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0170] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. An automatic vision-based loading and unloading method for a clutch-collaborating robot, characterized in that, The method includes: Based on the components to be loaded and unloaded corresponding to each process in the clutch production process, the form and position tolerance sensitive features of each component to be loaded and unloaded are extracted and simultaneously bound to each process node to generate a corresponding dynamic feature template library. When the loading and unloading command of the target process is triggered, the collaborative robot is driven to perform three-level positioning operations in sequence: global coarse positioning, local fine positioning, and tolerance feature verification, in order to determine the corresponding grasping pose. Simultaneously, combined with the cycle time requirements of the target process and the ready status of the upstream and downstream workstations on the production line, the corresponding initial loading and unloading path is generated. Based on the structure and material characteristics of the component corresponding to the target process, preset capture parameters are retrieved, and the initial loading and unloading path is optimized simultaneously based on the dynamic feature template library to generate the corresponding target loading and unloading path. Based on the preset grasping parameters, the collaborative robot is controlled to complete the corresponding grasping task according to the target loading and unloading path; The step of extracting the form and position tolerance sensitive features of each of the components to be loaded and unloaded corresponding to the process, and synchronously binding each process node to generate the corresponding dynamic feature template library includes: Multi-dimensional tolerance decomposition is performed on the components to be loaded and unloaded in each process of the clutch, so that the dimensional tolerance, position tolerance, and attitude tolerance are mapped into visually recognizable feature vectors respectively. Simultaneously, based on the processing risk level of the process, tolerance-sensitive weights are assigned to different feature vectors to eliminate redundant features and form lightweight feature primitives. The lightweight feature primitives are strongly associated with the tooling fixtures, gripping postures, and allowable deviation ranges of the corresponding processes to construct a ternary association structure of process-feature primitives-constraints. The feature drift thresholds under different batches and working conditions are recorded simultaneously to form an adaptive correction benchmark. Based on the ternary correlation structure and the adaptive correction benchmark, a dynamic feature template library that can be automatically loaded as the process switches is dynamically generated.

2. The automatic vision-based loading and unloading method for a clutch-cooperative robot according to claim 1, characterized in that, The step of dynamically generating the dynamic feature template library that can be automatically loaded during process switching based on the ternary correlation structure and the adaptive correction benchmark includes: Based on the ternary association structure, a process-feature template mapping index is constructed to assign a corresponding template identifier to each process node according to the mapping index. Simultaneously, the feature drift threshold in the adaptive correction benchmark is embedded into the template identifier to output the corresponding target feature data. Based on the target feature data, the feature primitives in the ternary association structure are compared with the adaptive correction benchmark to dynamically generate a template fit score. The optimal feature template is selected based on the template fit score, and the optimal feature template is dynamically updated simultaneously to generate the corresponding dynamic feature template library.

3. The automatic vision-based loading and unloading method for a clutch-cooperative robot according to claim 1, characterized in that, The step of generating the corresponding initial loading and unloading paths by combining the cycle time requirements of the target process and the readiness status of upstream and downstream workstations on the production line includes: The cycle time requirement of the target process is broken down into rigid time window constraints for the entire process of grabbing, transferring, and unloading. The actual ready status, preset action sequence and dynamic space occupancy information of the upstream and downstream workstations on the production line are collected simultaneously and transformed into dynamic space access constraints for path execution. Based on the dynamic spatial access constraints, combined with the grasping pose and the form and position tolerance sensitive features, a multi-dimensional coupled path feasible region is constructed. Simultaneously, based on the path feasible region, the entire loading and unloading process is decomposed into a grasping and positioning section, a smooth transfer section, and a precise unloading section. Differentiated constraint priorities are matched to the grabbing and positioning section, the smooth transfer section, and the precise unloading section. Simultaneously, the initial path planning is completed based on the timing matching of each constraint priority to generate the initial loading and unloading path.

4. The automatic vision-based loading and unloading method for a clutch-cooperative robot according to claim 3, characterized in that, The step of performing initial path planning based on the priority of each constraint to generate the initial loading and unloading path includes: Based on the respective constraint priorities, hierarchical and quantified constraint rules are constructed for the grasping and positioning segment, the smooth transfer segment, and the precise material release segment, respectively, to form a hierarchical constraint benchmark for path planning. Based on the aforementioned hierarchical constraint benchmark, a time-coupled segmented progressive planning algorithm is adopted to sequentially generate the initial path for each segment, and simultaneously set smooth transition constraint nodes at the junctions of adjacent segments. Each of the smooth transition constraint nodes is subjected to closed-loop verification to generate the corresponding initial loading and unloading path.

5. The automatic vision-based loading and unloading method for a clutch-cooperative robot according to claim 1, characterized in that, The step of optimizing the initial loading / unloading path based on the dynamic feature template library to generate the corresponding target loading / unloading path includes: In the dynamic feature template library, the form and position tolerance sensitive features, tolerance allowable thresholds and feature verification requirements of the component corresponding to the target process are retrieved, and combined with the preset capture parameters, they are converted into forced feature anchor points on the path; The spatial position, robot posture, and trigger timing of each of the forced feature anchor points are detected to integrate the corresponding initial optimization path. The initial optimized path is subjected to closed-loop checks for constraint compliance, cycle time adaptability, and spatial non-interference, in order to output the target loading and unloading path accordingly.

6. The automatic vision-based loading and unloading method for a clutch-cooperative robot according to claim 5, characterized in that, The step of performing closed-loop checks on the initial optimized path for constraint compliance, cycle time adaptability, and spatial non-interference to output the target loading / unloading path includes: Based on the dynamic feature template library, a digital twin verification model that maps to the actual physical production line is constructed. Simultaneously, the initial optimized path and the forced feature anchor points are fully mapped into the interior of the digital twin verification model to build a virtual verification carrier. Based on the virtual verification carrier, the initial optimized path is verified in a closed loop, and the deviation between the actual running data and the twin verification data is compared synchronously. The initial optimized path is reviewed and verified based on the deviation to generate the target loading and unloading path accordingly.

7. An automatic vision-based loading and unloading system for a clutch-collaborative robot, characterized in that, The system for implementing the automatic vision-based loading and unloading method for a clutch cooperative robot as described in any one of claims 1 to 6 includes: The extraction module is used to extract the form and position tolerance sensitive features of each process corresponding to the parts to be loaded and unloaded in each process of the clutch production process, and to bind each process node in a synchronous manner to generate a corresponding dynamic feature template library. The drive module is used to drive the collaborative robot to perform three levels of positioning operations in sequence: global coarse positioning, local fine positioning, and tolerance feature verification when the loading and unloading command of the target process is triggered, so as to determine the corresponding grasping posture and generate the corresponding initial loading and unloading path in combination with the cycle time requirements of the target process and the ready status of the upstream and downstream workstations on the production line. The processing module is used to retrieve preset capture parameters based on the structure and material characteristics of the component corresponding to the target process, and simultaneously optimize the initial loading and unloading path based on the dynamic feature template library to generate the corresponding target loading and unloading path. The control module is used to control the collaborative robot to complete the corresponding grasping task according to the target loading and unloading path based on the preset grasping parameters.

8. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the automatic vision loading and unloading method for a clutch cooperative robot as described in any one of claims 1 to 6.

9. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the automatic vision loading and unloading method for clutch cooperative robots as described in any one of claims 1 to 6.

Images